Susan K Halstead PhDGabriela Kalna PhDMohammad B Islam
MBBSIsrat Jahan MSCQuazi D Mohammad
MDBart C Jacobs MDHubert P Endtz MDZhahirul Islam PhDHugh J Willison
MBBS
Correspondence toProf WillisonHughWillisonglasgowacuk
Supplemental dataat Neurologyorgnn
Microarray screening of Guillain-Barreacutesyndrome sera for antibodies to glycolipidcomplexes
ABSTRACT
Objective To characterize the patterns of autoantibodies to glycolipid complexes in a large cohortof Guillain-Barreacute syndrome (GBS) and control samples collected in Bangladesh using a newlydeveloped microarray technique
Methods Twelve commonly studied glycolipids and lipids plus their 66 possible heteromericcomplexes totaling 78 antigens were applied to polyvinylidene fluoridendashcoated slides usinga microarray printer Arrays were probed with 266 GBS and 579 control sera (2 mL per serumdiluted 150) and bound immunoglobulin G detected with secondary antibody Scanned arrayswere subjected to statistical analyses
Results Measuring antibodies to single targets was 9 less sensitive than to heteromeric com-plex targets (492 vs 583) without significantly affecting specificity (839ndash850) Theoptimal screening protocol for GBS sera comprised a panel of 10 glycolipids (4 single glycolipidsGM1 GA1 GD1a GQ1b and their 6 heteromeric complexes) resulting in an overall assaysensitivity of 643 and specificity of 771 Notable heteromeric targets were GM1GD1aGM1GQ1b and GA1GD1a in which exclusive binding to the complex was observed
Conclusions Rationalizing the screening protocol to capture the enormous diversity of glycolipidcomplexes can be achieved by miniaturizing the screening platform to a microarray platform andapplying simple bioinformatics to determine optimal sensitivity and specificity of the targets Gly-colipid complexes are an important category of glycolipid antigens in autoimmune neuropathycases that require specific analytical and bioinformatics methods for optimal detection Neurol
Neuroimmunol Neuroinflamm 20163e284 doi 101212NXI0000000000000284
GLOSSARYAIDP 5 acute inflammatory demyelinating polyneuropathy AMAN 5 acute motor axonal neuropathy BSA 5 bovine serumalbumin FC 5 family controls GBS 5 Guillain-Barreacute syndrome IgG 5 immunoglobulin G NC 5 neurologic disease controlsPBS 5 phosphate-buffered saline ROC 5 receiver operating characteristic
Autoantibody binding to glycolipids that act as antigens in patients with autoimmune neurop-athy is heavily influenced by the topographic orientation of the carbohydrate head group withinliving peripheral nerve tissue and also within an immunoassay microenvironment1 Thusclusters of different lipids can interact to form complex molecular shapes capable of acting asantigens that are not detectable when assaying for individual glycolipid reactivities23 Equallysome antibody-binding sites on glycolipids may be obscured when the glycolipid is part ofa larger heteromeric lipid cluster This new category of glycolipid complexndashdependent autoanti-bodies has recently been described as either complex enhanced or complex attenuated Anti-glycolipid antibodies whose binding is unaffected by clustering are referred to as complexindependent
These authors are joint senior authors
From the Institute of Infection Immunity and Inflammation (SKH HJW) College of Medical Veterinary and Life Sciences University ofGlasgow The Beatson Institute for Cancer Research (GK) Glasgow UK Laboratory Sciences and Services Division (MBI IJ QDM ZI)International Centre for Diarrheal Disease Research Shaheed Tajuddin Ahmed Sarani (MBI IJ QDM ZI) Mohakhali Dhaka BangladeshDepartments of Immunology and Neurology (BCJ) and Medical Microbiology and Infectious Diseases (HPE) Erasmus MC UniversityMedical Center Rotterdam Rotterdam the Netherlands and Fondation Meacuterieux (HPE) Lyon France
Funding information and disclosures are provided at the end of the article Go to Neurologyorgnn for full disclosure forms The Article Processingcharge was paid by Editorial Office
This is an open access article distributed under the terms of the Creative Commons Attribution License 40 (CC BY) which permits unrestricteduse distribution and reproduction in any medium provided the original work is properly cited
Neurologyorgnn copy 2016 American Academy of Neurology 1
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Incorporating such findings into thedesign of screening assays for antiglycolipidantibodies adds substantial complexities towhat is already a difficult assay platform tostandardize even when using single glycoli-pids as antigens Thus if one considers 20 dif-ferent glycolipids as targets the number ofpossible heteromeric complexes in a 11 ratioamounts to 180 If one adds a third lipid tothe cluster or diversifies the ratios of thecluster components combinatorial complex-ity rises to unmanageable proportions whenusing routinely established ELISA-basedimmunoassays
To account for this and allow us to screenfor antibodies to highly varied glycolipidcomplexes in an unbiased way we havedeveloped a microanalytical method for as-saying cohorts of sera against multiplecombinatorial targets that advances our pre-viously reported methods In this proof ofprinciple study we selected 12 glycolipidsor lipids plus their 66 possible heteromericcomplexes totaling 78 antigens We appliedthis assay to identify previously reportedcombinatorial targets in a screen of 845Guillain-Barreacute syndrome (GBS) and controlsamples collected in Dhaka Bangladesh Weexpected that this geographical setting wouldprovide a high proportion of axonal GBSvariants and potentially a high chance of cap-turing antibodies to ganglioside complexesas have been previously observed in acutemotor axonal neuropathy (AMAN) Particu-larly in near-patient settings where early diag-nosis is useful simple biomarker testing kitscarrying high sensitivity and specificity thatcould be derived from these more complexdatasets are needed We report the findingsand identify the major clinically useful tar-gets in this patient group
METHODS Array fabrication For full methods please refer
to the e-Methods at Neurologyorgnn In brief array platforms
were produced in-house from polyvinylidene fluoride membrane
adhered to glass microscope slides Working solutions of single
glycolipids were prepared at 200 mgmL in methanol from which
heteromeric complexes were prepared Glycolipid samples were
stored at 220degC and sonicated prior to printing Glycolipids
microarray slides were produced using a Sciflexarrayer S3
microarray printer (Scienion Berlin Germany) A maximum of
20 slides was printed per run each containing 16 subarrays per
slide (320 arrays in total) All glycolipid targets were printed in
duplicate on each array and included methanol solvent which
was printed as a negative control Upon completion of printing
arrays were stored at 4degC until required
Clinical samples A total of 845 patients with GBS (with asso-
ciated clinical data) and control group sera were collected at Dha-
ka Medical College Hospital Bangladesh between 2010 and
2013 GBS cases were enrolled according to National Institute
of Neurological Disorders and Stroke criteria4 Samples com-
prised 266 patient sera (GBS) 258 family controls (FC) and
321 other neurologic disease controls (NC) Samples were stored
at 27080degC in the Laboratory Sciences and Services Division
All patients provided written informed consent the study was
approved by the ethics committees of the International Centre
for Diarrhoeal Disease Research Dhaka Bangladesh and Dhaka
Medical College Bangladesh
Sera screening Nonspecific serum binding was reduced by
blocking arrays in 2 bovine serum albumin (BSA)phosphate-
buffered saline (PBS) Each of the 16 subarrays per slide were iso-
lated using a FAST frame including a 16-well incubation
chamber (Maine Manufacturing Sanford ME) and 100 mL of
each serum sample diluted 150 in 1 BSAPBS was applied per
well for 1 hour at 4degC Samples were removed from each chamber
and then washed twice in 1 BSAPBS for 15 minutes at
room temperature Antibody binding was detected with
100 mL of 2 mgmL Alexafluor 647 conjugated goat antihuman
immunoglobulin G (IgG) (Jackson ImmunoResearchWest Grove
PA) per well for 1 hour at 4degC The arrays were then washed twice
in 1 BSAPBS for 30 minutes followed by twice in PBS for
5 minutes and a final 5 minutes wash in distilled water Each
serum sample was assayed in duplicate and screened twice in 2
independent assays
Scanning and analysis Arrays were scanned and quantitated
using a PerkinElmer (Waltham MA) ScanArray Express instru-
ment Image analysis was carried out with ProScanArray Express
Easy Quant software Each target spot was measured for median
fluorescence intensity with local background median pixel inten-
sity subtracted and the mean value was calculated for each pair of
duplicate spots Values obtained from repeat runs were averaged
and used in all calculations Data processing was performed with
Microsoft (Redmond WA) Excel Statistical analysis was per-
formed using censReg R GraphPad prism MedCalc and MeV
software Array data were compared with previously determined
ELISA data for GM1 positivity in 261266 of the GBS cases
and showed 889 agreement (positive or negative)
RESULTS Patient and control sample demographics
Suspected GBS cases recruited for this study (n 5
299) had the diagnosis of GBS confirmed on sub-sequent clinical evaluation in 266 patients A totalof 183 were male (688) and 83 were female(312) The medianmean age of patients withGBS was 2830 years (range 1ndash75) Electrophysio-logic examination at one time point after clinicalonset was performed on 192 patients (722) andclassified as follows acute inflammatory demyelinat-ing polyneuropathy (AIDP) 58 cases (302)AMAN 102 cases (531) unclassified GBS 32cases (167) Of the remaining 74 patients whodid not undergo electrophysiologic examination allbut one patient (who was classified as AIDP) werecategorized as unclassified GBS Patients reported the
2 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
following preceding symptoms (either in isolation orin combination with other symptoms) diarrhea 130cases (489) respiratory infection 59 cases(222) fever 34 cases (128) chicken pox 7cases (26) vaccination 6 cases (23) other 21cases (79) No preceding symptoms were reportedin 32 cases (120) There were 35 deaths (132)and 58 patients (218) required ventilation at somepoint during treatment For the family and neuro-logic control demographic and preceding symptomdata see the e-Methods
Antiglycolipid antibody screen The mean intensity val-ues were calculated from 4 intensity measurementsper target for each serum sample These 78 targetsare visually represented as a heat map employingPearson cluster coefficient and displayed as 3 groupsGBS (n 5 266) FC (n 5 258) and NC (n 5 321)with antibody binding intensity portrayed in a rain-bow scale red being high binding and blue being lowbinding (figure 1A) As expected for a GBS popula-tion in which multiple targets are known to be pres-ent no single antigen binding pattern was dominantthroughout the cohort examined Instead 30 targetswere found with significantly increased (p 005)binding intensities in GBS as compared indepen-dently with both FC and NC groups These dataon 30 significant targets are also displayed solely forthe GBS group for simplicity (figure 1B) Twenty-eight of the targets in the GBS group wereheteromeric complexes and 2 (GA1 and GM1)were single glycolipids While these 30 targets arehighly significantly different between disease andcontrol groups and may be clinically useful whenmultiple tests are performed screening for oneantigen or antigen complex cannot be used todetermine a measure of clinical usefulness byreceiver operating characteristic (ROC) analysis dueto the low sensitivity of individual targets Indeed asexpected when subjected to ROC analysis no targetsreached area under the curve $075 (the minimallevel considered to be the gold standard fora clinically useful biomarker5) when compared withcombined (family and neurologic) control groups
Certain targets (containing SGPG cholesteroland GA1) returned higher than average signals acrossboth disease and control groups indicating that nor-mal ranges need to be determined for each antigenOptimized positivity threshold values were calculatedfor each target antigen defined as the 95th percentileof the combined controls All intensity values wereconverted to binary data (positivenegative) deter-mined by the optimal threshold calculated for eachtarget and all subsequent data are expressed in termsof sensitivity and specificity When considering all 78targets a sensitivity of 827 and a specificity of
371 was achieved indicating that 629 of thecombined control samples had antibody bindingintensity greater than the threshold for one or moreof the 78 targets despite only 5 of controls beingpositive for any one individual antigen target In orderto increase the specificity of the assay we establisheda refined panel of target antigens
Anti-GM1 -GA1 -GD1a and -GQ1b antibodies Focus-ing on 4 glycolipids that are already recognized to betargets in GBS and were prominent antigens in thisscreen several common cluster patterns comprisingGM1 GA1 GD1a and GQ1b alone or in complexcould be identified (figure 2A) The largest group ofsera contained anti-GM1 antibodies (n 5 92sensitivity 5 346) Many of these also boundto GM1 when in heteromeric complex howeverin some cases binding to GM1 was inhibitedwhen in heteromeric complex with anotherglycolipid Considering the samples in whichheteromeric complexes enhanced anti-GM1antibody binding GM1GD1a and GM1GQ1bcomplexes were prominent pairings many of thesesamples containing completely complex-dependentantibodies (ie without the complex the antibodywould not be identifiable as there was no reactivityabove threshold for each individual ganglioside)Thus of the 98 (368) GBS sera that boundGM1GD1a complex 25 were completelycomplex dependent Of the 98 (368) GBS serathat bound GM1GQ1b complex 27 werecompletely complex dependent (figure 2 EndashF)
Antibody binding events were infrequent in theAMAN population for GA1 (37102 363) andGD1a (14102 137) when considered as singleantigens compared with other GBS populations inwhich anti-GD1a-positive AMAN cases are moreprominent6 However the GA1GD1a complex wasa significant target in 36 (353) AMAN samples ofwhich 10 were completely complex dependent (figure2 BndashD) Similarly GM1GD1a was a significant tar-get in 57 (559) AMAN samples of which 14 werecomplex dependent In this clinical population bothGA1 and GM1 therefore appear to enhance anti-GD1a antibody detection when the 2 glycolipidsare in complex
Complex enhancement and complex inhibition In orderto illustrate the enhancing and inhibiting effect ofcomplexes exemplary data are presented for GA1GD1a (figure 2 BndashD) and GM1GQ1b (figure 2EndashG) Heterogeneous patterns are found as ex-pected Thus there are 2 populations of anti-GA1GD1a antibodies in GBS sera (1) those inwhich binding intensities to GA1 or GD1a areattenuated by GA1GD1a complex and (2) thosein which binding is greatly enhanced by the
Neurology Neuroimmunology amp Neuroinflammation 3
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Figure 1 Heat map of antibody binding patterns to glycolipid targets in Guillain-Barreacute syndrome (GBS) casesand controls
(A) Patient (GBS) and control sera (family controls [FC] and neurologic disease controls [NC]) were screened against a panelof 78 single and heteromeric glycolipid targets on glycolipid microarray Mean fluorescent intensity values were calculatedand graphically presented as a heat map The rainbow scale indicates the intensity of the antibody binding in which blue isweak and red is strong antibody binding Pearson hierarchical clustering was employed to group samples with similar anti-gen binding patterns (B) Heat map presentation of GBS samples for the 30 glycolipid targets identified as having signifi-cantly different binding intensities between GBS and FC or NC groups (p 005)
4 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Figure 2 Subanalysis of 10 major antiglycolipid antibody targets in Guillain-Barreacute syndrome (GBS) cases
(A) Heat map presentation of patient (GBS) and control sera (family controls [FC] and neurologic disease controls [NC]) reactivity with GM1 GD1a GQ1b andGA1 as single antigens and their 6 possible heteromeric complexes When selecting this small panel of 10 targets as serum biomarkers of GBS in this clinicalcohort the combined sensitivity is 643 and the specificity is 771 (B) Dot plot presentation of GBS (n5266) antibody binding intensities for single GA1and GD1a and their heteromeric complex GA1GD1a (C) Heat map presentation of patient binding intensities Each row represents a single patient withbinding intensity to each target represented using the rainbow scale (D) Line graph compares the binding intensity for complex GA1GD1a with the sum ofthe single antigens (GA1 and GD1a) Green lines indicates complex enhancement while red lines represent complex attenuation (E) Dot plot presentation ofGBS (n 5 266) antibody binding intensities for single GM1 and GQ1b and their heteromeric complex GM1GQ1b (F) Heat map presentation of patientbinding intensities Each row represents a single patient with binding to each target represented using the rainbow scale (G) Line graph compares thebinding intensity for complex GM1GQ1b with the sum of the single antigens (GM1 and GQ1b) Green lines indicates complex enhancement while red linesrepresent complex attenuation
Neurology Neuroimmunology amp Neuroinflammation 5
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
complex most usually in a completely complex-dependent fashion (ie no binding is present toeither partner alone figure 2D green lines)Similarly when considering GM1 and GQ1bwhile some anti-GM1 and anti-GQ1b sera haveattenuated binding intensities when in GM1GQ1b complex there is a significantly largegroup of sera in which major enhancement
occurs in the presence of the complex (figure 2Ggreen lines)
The added value and limitations of screening for
heteromeric complexes To assess the overall impactof including glycolipid heteromeric complexes inthe screening platform we compared the number ofGBS samples classified as positive (ie returning anintensity value 95th percentile of the combinedcontrols) for single glycolipids and heteromeric com-plexes for the 10 frequently observed reactivities (fig-ure 2A) A total of 131 of 266 GBS samples and 87 of579 combined controls were positive for one or moresingle targets (sensitivity 492 specificity 850)whereas 155266 GBS samples and 95579 com-bined controls were positive for one or more hetero-meric complexes (sensitivity 583 specificity836) Therefore by screening for heteromericcomplexes a gain of 91 in sensitivity (p 5
00021) is achieved without a significant loss in spec-ificity (14 p 5 0445) Screening of GBS serairrespective of the clinical variant (AMAN AIDPor unclassified) against this discrete panel of 10 gly-colipids (4 single and 6 complexes) resulted in anoverall assay sensitivity of 643 and specificity of771
However when examining a larger panel of tar-gets a gain in sensitivity is frequently offset by anequivalent or greater loss in specificity (figure 3A)For example when examining the 30 targets withsignificantly different fluorescent intensity values (fig-ure 1B) a gain of 304 in sensitivity (from 410to 714) is offset by a loss of 29 specificity (from912 to 622) when comparing single and com-plex positivity frequency Similarly when consideringall 78 targets printed on the glycolipid arrays there isa 147 gain in sensitivity (from 650 to 797)while specificity drops 243 (from 658 to415) for single and complexes respectively Thesedata show that when examining the data in its total-ity the disease specificity of a very broad unbiasedscreen is expectedly poor owing to the large numberof controls whose sera contain at least one species ofantiglycolipid antibody above the normal range(95th percentile)
Antiglycolipid antibodies associated with clinical
variants Correlation between antiglycolipid anti-body profiles and clinical features are presented intable and as e-Results Having determined a panelof 10 glycolipid reactivities optimized for sensitivityand specificity for all patients with GBS irrespectiveof the clinical variant next we sought to determinewhether the presence of specific antibody reactiv-ities would segregate with a clinical variant or couldbe used to predict the clinical phenotype or out-come (see e-Results) Patients who had unclassified
Figure 3 Sensitivity and specificity of microarray analysis in relation to targetnumber
(A) Comparison of overall assay accuracy (both sensitivity and specificity) when screeningagainst an increasing number of target antigens Assay specificity declines as the numbersof target antigens increases Improved assay performance is achieved when a small panel of10 targets is selected (B) Dot plot compares the maximum binding intensity signal for eachpatient or control when comparing single and heteromeric complexes GBS patient medianarbitrary binding intensity for single targets (7473 FIU) is significantly less than for hetero-meric complex targets (35899 FIU p 00001) indicating that heteromeric complexesreturn higher signals than single antigens
6 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
GBS variants (n 5 105) were excluded from thisanalysis Six targets were identified as being signif-icant in the AIDP population when compared withthe AMAN clinical variant These included SGPGalone as well as 2 complexes containing SGPG(SGPGGQ1b and SGPGChol) In addition 2heteromeric complexes containing LM1 (LM1Sul-phatide and LM1Chol) and CTHChol were sig-nificantly associated with AIDP (p 005) Whilesignificant each of these targets was present at low
frequencies within the patient samples and thusindividually they had low diagnostic sensitivity(ranging from 105 to 140) but with a highspecificity both for combined controls (950) andpatients with AMAN (ranging from 960 to980 depending upon the target) When com-bined as a diagnostic panel of 6 antigen targets forAIDP a sensitivity of 339 and a specificity of873 and 834 (for AMAN and controls respec-tively) was reached
Table Glycolipid reactivities associated with acute inflammatory demyelinating polyneuropathy (AIDP)andacute motor axonal neuropathy (AMAN) clinical variants
AIDP (n 5 59) AMAN (n 5 102)
Guillain-Barreacute syndrome variant biomarker SGPG GM1 single 1 All GM1 complexes
SGPGGQ1b GA1 single 1 All GA1 complexes (excl GA1GQ1b)
SGPGChol GD1bGD1a
LM1Sulph GD1bPS
LM1Chol GD1bChol
CTHChol GD1bSulph
Ocular deficits GQ1bGD1b NA
GQ1bSGPG
GQ1bCTH
GQ1bChol
GQ1bGalC
LM1CTH
Bulbar palsy NA GQ1bPS
Facial palsy NA GA1GD1a
Figure 4 Heat map of antibody binding patterns in axonal vs demyelinating Guillain-Barreacute syndrome (GBS)cases
(A) Heat map presentation of the 26 glycolipid targets significant for the acute motor axonal neuropathy variant of GBScompared with the demyelinating (acute inflammatory demyelinating polyneuropathy) clinical subtype (B) Refining thescreening targets to a panel of 3 heteromeric complexes (GM1GQ1b GM1phosphatidylserine and GA1GD1a) resultsin a sensitivity of 784 and a specificity of 877 (for combined controls)
Neurology Neuroimmunology amp Neuroinflammation 7
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Twenty-six targets were identified as being sig-nificant (p 005) in AMAN when compared withthe AIDP population (figure 4A) These includeGM1 alone plus GM1 in complex with all testedlipids (GM1GA1 GM1GD1a GM1GD1bGM1GQ1b GM1SGPG GM1LM1 GM1CTH GM1PS GM1Chol GM1SulphatideGM1GalC) GA1 alone plus GA1 in complex withall tested lipids with the exception of GA1GQ1b(GA1GD1a GA1GD1b GA1SGPG GA1LM1GA1CTH GA1PS GA1Chol GA1SulphatideGA1GalC) GD1bGD1a GD1bPS GD1bCholand GD1bSulphatide Of these GM1GD1a hadthe highest sensitivity (559) and a high specific-ity for both combined controls (950) and pa-tients with AIDP (848) When combined asa diagnostic panel of 26 antigen targets for AMANa sensitivity of 863 and a specificity of 644and 458 (for combined controls and AIDPrespectively) was reached As previously demon-strated the specificity was reduced with increasingantigen targets therefore we sought to refine thenumber of antigen targets in order to optimize sen-sitivity and specificity When selecting only 3 anti-gen targets (GM1GQ1b GM1PS and GA1GD1a) a sensitivity of 784 and specificity of877 and 610 (for combined control andAIDP respectively) was achieved (figure 4B)
DISCUSSION This study applies technical develop-ments in antiglycolipid antibody assay miniaturizationto address the practical complexities of large-scalebiomarker screening of clinical cohorts By adaptingthe microarray printer to accommodate glycolipidswe were able to print arrays in a high throughputmanner One maximum capacity assay is capable ofprinting 320 individual arrays on a total of 20slides Conducting an equivalent screen on thissample size and antigen repertoire in a conventionalELISA study would require 640 ELISA plates (96-well) Miniaturization of the glycolipid assay allowsus to use 100-fold smaller volumes of patient serumthus the total serum use by microarray is 2 mL toprobe against 78 target antigens with duplicatemeasurement compared with 200 mL of serum fora comparable study performed on ELISA In additionthe use of fluorescently conjugated antibodies allowsfor multiplex screening whereby IgG andimmunoglobulin M can be detected simultaneouslyon each assay platform a benefit not routinelyavailable in ELISA readers
In this study a single electrophysiologic assess-ment was performed on 192 of the 266 patientsin the GBS cohort Of these patients 531 of pa-tients were classified as having an axonal variant ofGBS 302 of patients with AIDP while the
clinical variant of the remaining 167 of patientswas unclassified These diagnostic categorizationfrequencies are in line with a previously reportedGBS cohort from Bangladesh7 As found here thisprevious report also identified a low incidence ofpatient serum containing anti-GD1a antibodies(14) which was surprising given the high fre-quency of axonal GBS in comparison with studiesfrom elsewhere68ndash10 However probing Bangladeshsera against heteromeric complexes provedadvantageous as it revealed populations of complex-dependent GA1GD1a and GM1GD1a antibodieswhich were distinct from antibodies binding GD1a asa single antigen Similarly GM1GQ1b complex-dependent antibody binding was identified as a dis-tinct population in 119 of AMAN samples whencompared to GM1 and GQ1b alone Using a refinedpanel of only 3 antiglycolipid targets a high sensitivityand specificity was achieved for patients with AMANin this cohort
In addition to screening GBS patient samples 579combined control sera were examined This screenidentified variations in the baseline levels of antigan-glioside antibodies reactivities for each of the targetsWhen performing conventional ELISA the currentgold standard method for antiganglioside antibodydetection a universal optical density threshold of pos-itivity across all glycolipids is widely used In thisgeographically restricted study we can see that inorder to optimize both sensitivity and specificity ofthe assay individual threshold must be calculatedin order to determine the normal range of these nat-urally occurring and GBS-independent anticarbohy-drate antibodies
In this study we present data that identify thepresence of both single and heteromeric glycolipidcomplex binding antibodies in this large GBS cohortInitial biomarker screening employed a panel of 78targets however the selective inclusion of a very lim-ited panel of targets enabled optimization of both sen-sitivity and specificity due to significant targetredundancy resulting from the presence of polyclonalantibodies or cross-reactivity of specific antibodyspecies When considering the overwhelming com-plexities of identifying diagnostically important glyco-lipid complexes by ELISA screening the necessitiesof miniaturization are clearly evident for the initialanalysis in order to find the markers that are mostinformative
AUTHOR CONTRIBUTIONSSK Halstead study concept and design experimental work principal
draft of manuscript G Kalna statistical analysis and interpretation
MB Islam acquisition of clinical data I Jahan acquisition of clinical
data QD Mohammad acquisition of clinical data BC Jacobs study
concept and design study supervision HP Endtz study concept and
design study supervision Z Islam study concept and design critical
8 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
revision of the manuscript for important intellectual content HJ Willison
study concept and design critical revision of the manuscript for important
intellectual content study supervision
STUDY FUNDINGThis study is funded by the Wellcome Trust and the Chandra Mehta
Foundation
DISCLOSURESK Halstead G Kalna MB Islam I Jahan and QD Mohammad
report no disclosures BC Jacobs received travel funding from Baxter
International is on the editorial board for Journal of the Peripheral Ner-
vous System and received research support from Grifols CSL-Behring
Netherlands Organization for Health Research and Development Eras-
mus MC Prinses Beatrix GBS-CIDP Foundation International Prinses
Beatrix Fonds and Prinses Beatrix Spierfonds HP Endtz received
research support from EC Horizon 2020 Foundation Merieux Bill amp
Melinda Gates Foundation and GBS-CIDP foundation Z Islam reports
no disclosures HJ Willison is on the editorial board for Nature Clinical
Practice Neurology Journal of Neuroimmunology Muscle and Nerve Jour-
nal of Peripheral Nervous System Clinical and Experimental Neuroimmu-
nology and Experimental Neurology is an associate editor for BMC
Neurology holds a patent for combinatorial glycoarray technology and
received research support from Ipsen Annexon Alexion and MRC Go
to Neurologyorgnn for full disclosure forms
Received April 4 2016 Accepted in final form August 5 2016
REFERENCES1 Rinaldi S Brennan KM Willison HJ Heteromeric glyco-
lipid complexes as modulators of autoantibody and lectin
binding Prog Lipid Res 20104987ndash95
2 Kaida K Morita D Kanzaki M et al Ganglioside com-
plexes as new target antigens in Guillain-Barre syndrome
Ann Neurol 200456567ndash571
3 Galban-Horcajo F Halstead SK McGonigal R
Willison HJ The application of glycosphingolipid arrays
to autoantibody detection in neuroimmunological disor-
ders Curr Opin Chem Biol 20141878ndash86
4 Asbury AK Cornblath DR Assessment of current diag-
nostic criteria for Guillain-Barre syndrome Ann Neurol
1990(27 suppl)S21ndashS24
5 Jones CM Athanasiou T Summary receiver operating
characteristic curve analysis techniques in the evaluation
of diagnostic tests Ann Thorac Surg 20057916ndash20
6 Ho TW Willison HJ Nachamkin I et al Anti-GD1a
antibody is associated with axonal but not demyelinating
forms of Guillain-Barre syndrome Ann Neurol 199945
168ndash173
7 Islam Z Jacobs BC van Belkum A et al Axonal var-
iant of Guillain-Barre syndrome associated with cam-
pylobacter infection in Bangladesh Neurology 2010
74581ndash587
8 Kim JK Bae JS Kim DS et al Prevalence of anti-
ganglioside antibodies and their clinical correlates with
Guillain-Barreacute syndrome in Korea a nationwide multicen-
ter study J Clin Neurol 20141094ndash100
9 Willison HJ Yuki N Peripheral neuropathies and anti-
glycolipid antibodies Brain 20021252591ndash2625
10 Yuki N Yamada M Sato S et al Association of IgG anti-
GD1a antibody with severe Guillain-Barre syndrome
Muscle Nerve 199316642ndash647
Neurology Neuroimmunology amp Neuroinflammation 9
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
DOI 101212NXI000000000000028420163 Neurol Neuroimmunol Neuroinflamm
Susan K Halstead Gabriela Kalna Mohammad B Islam et al complexes
Microarray screening of Guillain-Barreacute syndrome sera for antibodies to glycolipid
This information is current as of September 28 2016
ServicesUpdated Information amp
httpnnneurologyorgcontent36e284fullhtmlincluding high resolution figures can be found at
Supplementary Material
httpnnneurologyorgcontentsuppl2017022136e284DC2 httpnnneurologyorgcontentsuppl2016092836e284DC1
Supplementary material can be found at
References httpnnneurologyorgcontent36e284fullhtmlref-list-1
This article cites 9 articles 0 of which you can access for free at
Subspecialty Collections
httpnnneurologyorgcgicollectionperipheral_neuropathyPeripheral neuropathy
httpnnneurologyorgcgicollectionguillainbarre_syndromeGuillain-Barre syndrome
httpnnneurologyorgcgicollectionautoimmune_diseasesAutoimmune diseasesfollowing 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
2016 American Academy of Neurology All rights reserved Online ISSN 2332-7812Published since April 2014 it is an open-access online-only continuous publication journal Copyright copy
is an official journal of the American Academy of NeurologyNeurol Neuroimmunol Neuroinflamm
Incorporating such findings into thedesign of screening assays for antiglycolipidantibodies adds substantial complexities towhat is already a difficult assay platform tostandardize even when using single glycoli-pids as antigens Thus if one considers 20 dif-ferent glycolipids as targets the number ofpossible heteromeric complexes in a 11 ratioamounts to 180 If one adds a third lipid tothe cluster or diversifies the ratios of thecluster components combinatorial complex-ity rises to unmanageable proportions whenusing routinely established ELISA-basedimmunoassays
To account for this and allow us to screenfor antibodies to highly varied glycolipidcomplexes in an unbiased way we havedeveloped a microanalytical method for as-saying cohorts of sera against multiplecombinatorial targets that advances our pre-viously reported methods In this proof ofprinciple study we selected 12 glycolipidsor lipids plus their 66 possible heteromericcomplexes totaling 78 antigens We appliedthis assay to identify previously reportedcombinatorial targets in a screen of 845Guillain-Barreacute syndrome (GBS) and controlsamples collected in Dhaka Bangladesh Weexpected that this geographical setting wouldprovide a high proportion of axonal GBSvariants and potentially a high chance of cap-turing antibodies to ganglioside complexesas have been previously observed in acutemotor axonal neuropathy (AMAN) Particu-larly in near-patient settings where early diag-nosis is useful simple biomarker testing kitscarrying high sensitivity and specificity thatcould be derived from these more complexdatasets are needed We report the findingsand identify the major clinically useful tar-gets in this patient group
METHODS Array fabrication For full methods please refer
to the e-Methods at Neurologyorgnn In brief array platforms
were produced in-house from polyvinylidene fluoride membrane
adhered to glass microscope slides Working solutions of single
glycolipids were prepared at 200 mgmL in methanol from which
heteromeric complexes were prepared Glycolipid samples were
stored at 220degC and sonicated prior to printing Glycolipids
microarray slides were produced using a Sciflexarrayer S3
microarray printer (Scienion Berlin Germany) A maximum of
20 slides was printed per run each containing 16 subarrays per
slide (320 arrays in total) All glycolipid targets were printed in
duplicate on each array and included methanol solvent which
was printed as a negative control Upon completion of printing
arrays were stored at 4degC until required
Clinical samples A total of 845 patients with GBS (with asso-
ciated clinical data) and control group sera were collected at Dha-
ka Medical College Hospital Bangladesh between 2010 and
2013 GBS cases were enrolled according to National Institute
of Neurological Disorders and Stroke criteria4 Samples com-
prised 266 patient sera (GBS) 258 family controls (FC) and
321 other neurologic disease controls (NC) Samples were stored
at 27080degC in the Laboratory Sciences and Services Division
All patients provided written informed consent the study was
approved by the ethics committees of the International Centre
for Diarrhoeal Disease Research Dhaka Bangladesh and Dhaka
Medical College Bangladesh
Sera screening Nonspecific serum binding was reduced by
blocking arrays in 2 bovine serum albumin (BSA)phosphate-
buffered saline (PBS) Each of the 16 subarrays per slide were iso-
lated using a FAST frame including a 16-well incubation
chamber (Maine Manufacturing Sanford ME) and 100 mL of
each serum sample diluted 150 in 1 BSAPBS was applied per
well for 1 hour at 4degC Samples were removed from each chamber
and then washed twice in 1 BSAPBS for 15 minutes at
room temperature Antibody binding was detected with
100 mL of 2 mgmL Alexafluor 647 conjugated goat antihuman
immunoglobulin G (IgG) (Jackson ImmunoResearchWest Grove
PA) per well for 1 hour at 4degC The arrays were then washed twice
in 1 BSAPBS for 30 minutes followed by twice in PBS for
5 minutes and a final 5 minutes wash in distilled water Each
serum sample was assayed in duplicate and screened twice in 2
independent assays
Scanning and analysis Arrays were scanned and quantitated
using a PerkinElmer (Waltham MA) ScanArray Express instru-
ment Image analysis was carried out with ProScanArray Express
Easy Quant software Each target spot was measured for median
fluorescence intensity with local background median pixel inten-
sity subtracted and the mean value was calculated for each pair of
duplicate spots Values obtained from repeat runs were averaged
and used in all calculations Data processing was performed with
Microsoft (Redmond WA) Excel Statistical analysis was per-
formed using censReg R GraphPad prism MedCalc and MeV
software Array data were compared with previously determined
ELISA data for GM1 positivity in 261266 of the GBS cases
and showed 889 agreement (positive or negative)
RESULTS Patient and control sample demographics
Suspected GBS cases recruited for this study (n 5
299) had the diagnosis of GBS confirmed on sub-sequent clinical evaluation in 266 patients A totalof 183 were male (688) and 83 were female(312) The medianmean age of patients withGBS was 2830 years (range 1ndash75) Electrophysio-logic examination at one time point after clinicalonset was performed on 192 patients (722) andclassified as follows acute inflammatory demyelinat-ing polyneuropathy (AIDP) 58 cases (302)AMAN 102 cases (531) unclassified GBS 32cases (167) Of the remaining 74 patients whodid not undergo electrophysiologic examination allbut one patient (who was classified as AIDP) werecategorized as unclassified GBS Patients reported the
2 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
following preceding symptoms (either in isolation orin combination with other symptoms) diarrhea 130cases (489) respiratory infection 59 cases(222) fever 34 cases (128) chicken pox 7cases (26) vaccination 6 cases (23) other 21cases (79) No preceding symptoms were reportedin 32 cases (120) There were 35 deaths (132)and 58 patients (218) required ventilation at somepoint during treatment For the family and neuro-logic control demographic and preceding symptomdata see the e-Methods
Antiglycolipid antibody screen The mean intensity val-ues were calculated from 4 intensity measurementsper target for each serum sample These 78 targetsare visually represented as a heat map employingPearson cluster coefficient and displayed as 3 groupsGBS (n 5 266) FC (n 5 258) and NC (n 5 321)with antibody binding intensity portrayed in a rain-bow scale red being high binding and blue being lowbinding (figure 1A) As expected for a GBS popula-tion in which multiple targets are known to be pres-ent no single antigen binding pattern was dominantthroughout the cohort examined Instead 30 targetswere found with significantly increased (p 005)binding intensities in GBS as compared indepen-dently with both FC and NC groups These dataon 30 significant targets are also displayed solely forthe GBS group for simplicity (figure 1B) Twenty-eight of the targets in the GBS group wereheteromeric complexes and 2 (GA1 and GM1)were single glycolipids While these 30 targets arehighly significantly different between disease andcontrol groups and may be clinically useful whenmultiple tests are performed screening for oneantigen or antigen complex cannot be used todetermine a measure of clinical usefulness byreceiver operating characteristic (ROC) analysis dueto the low sensitivity of individual targets Indeed asexpected when subjected to ROC analysis no targetsreached area under the curve $075 (the minimallevel considered to be the gold standard fora clinically useful biomarker5) when compared withcombined (family and neurologic) control groups
Certain targets (containing SGPG cholesteroland GA1) returned higher than average signals acrossboth disease and control groups indicating that nor-mal ranges need to be determined for each antigenOptimized positivity threshold values were calculatedfor each target antigen defined as the 95th percentileof the combined controls All intensity values wereconverted to binary data (positivenegative) deter-mined by the optimal threshold calculated for eachtarget and all subsequent data are expressed in termsof sensitivity and specificity When considering all 78targets a sensitivity of 827 and a specificity of
371 was achieved indicating that 629 of thecombined control samples had antibody bindingintensity greater than the threshold for one or moreof the 78 targets despite only 5 of controls beingpositive for any one individual antigen target In orderto increase the specificity of the assay we establisheda refined panel of target antigens
Anti-GM1 -GA1 -GD1a and -GQ1b antibodies Focus-ing on 4 glycolipids that are already recognized to betargets in GBS and were prominent antigens in thisscreen several common cluster patterns comprisingGM1 GA1 GD1a and GQ1b alone or in complexcould be identified (figure 2A) The largest group ofsera contained anti-GM1 antibodies (n 5 92sensitivity 5 346) Many of these also boundto GM1 when in heteromeric complex howeverin some cases binding to GM1 was inhibitedwhen in heteromeric complex with anotherglycolipid Considering the samples in whichheteromeric complexes enhanced anti-GM1antibody binding GM1GD1a and GM1GQ1bcomplexes were prominent pairings many of thesesamples containing completely complex-dependentantibodies (ie without the complex the antibodywould not be identifiable as there was no reactivityabove threshold for each individual ganglioside)Thus of the 98 (368) GBS sera that boundGM1GD1a complex 25 were completelycomplex dependent Of the 98 (368) GBS serathat bound GM1GQ1b complex 27 werecompletely complex dependent (figure 2 EndashF)
Antibody binding events were infrequent in theAMAN population for GA1 (37102 363) andGD1a (14102 137) when considered as singleantigens compared with other GBS populations inwhich anti-GD1a-positive AMAN cases are moreprominent6 However the GA1GD1a complex wasa significant target in 36 (353) AMAN samples ofwhich 10 were completely complex dependent (figure2 BndashD) Similarly GM1GD1a was a significant tar-get in 57 (559) AMAN samples of which 14 werecomplex dependent In this clinical population bothGA1 and GM1 therefore appear to enhance anti-GD1a antibody detection when the 2 glycolipidsare in complex
Complex enhancement and complex inhibition In orderto illustrate the enhancing and inhibiting effect ofcomplexes exemplary data are presented for GA1GD1a (figure 2 BndashD) and GM1GQ1b (figure 2EndashG) Heterogeneous patterns are found as ex-pected Thus there are 2 populations of anti-GA1GD1a antibodies in GBS sera (1) those inwhich binding intensities to GA1 or GD1a areattenuated by GA1GD1a complex and (2) thosein which binding is greatly enhanced by the
Neurology Neuroimmunology amp Neuroinflammation 3
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Figure 1 Heat map of antibody binding patterns to glycolipid targets in Guillain-Barreacute syndrome (GBS) casesand controls
(A) Patient (GBS) and control sera (family controls [FC] and neurologic disease controls [NC]) were screened against a panelof 78 single and heteromeric glycolipid targets on glycolipid microarray Mean fluorescent intensity values were calculatedand graphically presented as a heat map The rainbow scale indicates the intensity of the antibody binding in which blue isweak and red is strong antibody binding Pearson hierarchical clustering was employed to group samples with similar anti-gen binding patterns (B) Heat map presentation of GBS samples for the 30 glycolipid targets identified as having signifi-cantly different binding intensities between GBS and FC or NC groups (p 005)
4 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Figure 2 Subanalysis of 10 major antiglycolipid antibody targets in Guillain-Barreacute syndrome (GBS) cases
(A) Heat map presentation of patient (GBS) and control sera (family controls [FC] and neurologic disease controls [NC]) reactivity with GM1 GD1a GQ1b andGA1 as single antigens and their 6 possible heteromeric complexes When selecting this small panel of 10 targets as serum biomarkers of GBS in this clinicalcohort the combined sensitivity is 643 and the specificity is 771 (B) Dot plot presentation of GBS (n5266) antibody binding intensities for single GA1and GD1a and their heteromeric complex GA1GD1a (C) Heat map presentation of patient binding intensities Each row represents a single patient withbinding intensity to each target represented using the rainbow scale (D) Line graph compares the binding intensity for complex GA1GD1a with the sum ofthe single antigens (GA1 and GD1a) Green lines indicates complex enhancement while red lines represent complex attenuation (E) Dot plot presentation ofGBS (n 5 266) antibody binding intensities for single GM1 and GQ1b and their heteromeric complex GM1GQ1b (F) Heat map presentation of patientbinding intensities Each row represents a single patient with binding to each target represented using the rainbow scale (G) Line graph compares thebinding intensity for complex GM1GQ1b with the sum of the single antigens (GM1 and GQ1b) Green lines indicates complex enhancement while red linesrepresent complex attenuation
Neurology Neuroimmunology amp Neuroinflammation 5
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
complex most usually in a completely complex-dependent fashion (ie no binding is present toeither partner alone figure 2D green lines)Similarly when considering GM1 and GQ1bwhile some anti-GM1 and anti-GQ1b sera haveattenuated binding intensities when in GM1GQ1b complex there is a significantly largegroup of sera in which major enhancement
occurs in the presence of the complex (figure 2Ggreen lines)
The added value and limitations of screening for
heteromeric complexes To assess the overall impactof including glycolipid heteromeric complexes inthe screening platform we compared the number ofGBS samples classified as positive (ie returning anintensity value 95th percentile of the combinedcontrols) for single glycolipids and heteromeric com-plexes for the 10 frequently observed reactivities (fig-ure 2A) A total of 131 of 266 GBS samples and 87 of579 combined controls were positive for one or moresingle targets (sensitivity 492 specificity 850)whereas 155266 GBS samples and 95579 com-bined controls were positive for one or more hetero-meric complexes (sensitivity 583 specificity836) Therefore by screening for heteromericcomplexes a gain of 91 in sensitivity (p 5
00021) is achieved without a significant loss in spec-ificity (14 p 5 0445) Screening of GBS serairrespective of the clinical variant (AMAN AIDPor unclassified) against this discrete panel of 10 gly-colipids (4 single and 6 complexes) resulted in anoverall assay sensitivity of 643 and specificity of771
However when examining a larger panel of tar-gets a gain in sensitivity is frequently offset by anequivalent or greater loss in specificity (figure 3A)For example when examining the 30 targets withsignificantly different fluorescent intensity values (fig-ure 1B) a gain of 304 in sensitivity (from 410to 714) is offset by a loss of 29 specificity (from912 to 622) when comparing single and com-plex positivity frequency Similarly when consideringall 78 targets printed on the glycolipid arrays there isa 147 gain in sensitivity (from 650 to 797)while specificity drops 243 (from 658 to415) for single and complexes respectively Thesedata show that when examining the data in its total-ity the disease specificity of a very broad unbiasedscreen is expectedly poor owing to the large numberof controls whose sera contain at least one species ofantiglycolipid antibody above the normal range(95th percentile)
Antiglycolipid antibodies associated with clinical
variants Correlation between antiglycolipid anti-body profiles and clinical features are presented intable and as e-Results Having determined a panelof 10 glycolipid reactivities optimized for sensitivityand specificity for all patients with GBS irrespectiveof the clinical variant next we sought to determinewhether the presence of specific antibody reactiv-ities would segregate with a clinical variant or couldbe used to predict the clinical phenotype or out-come (see e-Results) Patients who had unclassified
Figure 3 Sensitivity and specificity of microarray analysis in relation to targetnumber
(A) Comparison of overall assay accuracy (both sensitivity and specificity) when screeningagainst an increasing number of target antigens Assay specificity declines as the numbersof target antigens increases Improved assay performance is achieved when a small panel of10 targets is selected (B) Dot plot compares the maximum binding intensity signal for eachpatient or control when comparing single and heteromeric complexes GBS patient medianarbitrary binding intensity for single targets (7473 FIU) is significantly less than for hetero-meric complex targets (35899 FIU p 00001) indicating that heteromeric complexesreturn higher signals than single antigens
6 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
GBS variants (n 5 105) were excluded from thisanalysis Six targets were identified as being signif-icant in the AIDP population when compared withthe AMAN clinical variant These included SGPGalone as well as 2 complexes containing SGPG(SGPGGQ1b and SGPGChol) In addition 2heteromeric complexes containing LM1 (LM1Sul-phatide and LM1Chol) and CTHChol were sig-nificantly associated with AIDP (p 005) Whilesignificant each of these targets was present at low
frequencies within the patient samples and thusindividually they had low diagnostic sensitivity(ranging from 105 to 140) but with a highspecificity both for combined controls (950) andpatients with AMAN (ranging from 960 to980 depending upon the target) When com-bined as a diagnostic panel of 6 antigen targets forAIDP a sensitivity of 339 and a specificity of873 and 834 (for AMAN and controls respec-tively) was reached
Table Glycolipid reactivities associated with acute inflammatory demyelinating polyneuropathy (AIDP)andacute motor axonal neuropathy (AMAN) clinical variants
AIDP (n 5 59) AMAN (n 5 102)
Guillain-Barreacute syndrome variant biomarker SGPG GM1 single 1 All GM1 complexes
SGPGGQ1b GA1 single 1 All GA1 complexes (excl GA1GQ1b)
SGPGChol GD1bGD1a
LM1Sulph GD1bPS
LM1Chol GD1bChol
CTHChol GD1bSulph
Ocular deficits GQ1bGD1b NA
GQ1bSGPG
GQ1bCTH
GQ1bChol
GQ1bGalC
LM1CTH
Bulbar palsy NA GQ1bPS
Facial palsy NA GA1GD1a
Figure 4 Heat map of antibody binding patterns in axonal vs demyelinating Guillain-Barreacute syndrome (GBS)cases
(A) Heat map presentation of the 26 glycolipid targets significant for the acute motor axonal neuropathy variant of GBScompared with the demyelinating (acute inflammatory demyelinating polyneuropathy) clinical subtype (B) Refining thescreening targets to a panel of 3 heteromeric complexes (GM1GQ1b GM1phosphatidylserine and GA1GD1a) resultsin a sensitivity of 784 and a specificity of 877 (for combined controls)
Neurology Neuroimmunology amp Neuroinflammation 7
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Twenty-six targets were identified as being sig-nificant (p 005) in AMAN when compared withthe AIDP population (figure 4A) These includeGM1 alone plus GM1 in complex with all testedlipids (GM1GA1 GM1GD1a GM1GD1bGM1GQ1b GM1SGPG GM1LM1 GM1CTH GM1PS GM1Chol GM1SulphatideGM1GalC) GA1 alone plus GA1 in complex withall tested lipids with the exception of GA1GQ1b(GA1GD1a GA1GD1b GA1SGPG GA1LM1GA1CTH GA1PS GA1Chol GA1SulphatideGA1GalC) GD1bGD1a GD1bPS GD1bCholand GD1bSulphatide Of these GM1GD1a hadthe highest sensitivity (559) and a high specific-ity for both combined controls (950) and pa-tients with AIDP (848) When combined asa diagnostic panel of 26 antigen targets for AMANa sensitivity of 863 and a specificity of 644and 458 (for combined controls and AIDPrespectively) was reached As previously demon-strated the specificity was reduced with increasingantigen targets therefore we sought to refine thenumber of antigen targets in order to optimize sen-sitivity and specificity When selecting only 3 anti-gen targets (GM1GQ1b GM1PS and GA1GD1a) a sensitivity of 784 and specificity of877 and 610 (for combined control andAIDP respectively) was achieved (figure 4B)
DISCUSSION This study applies technical develop-ments in antiglycolipid antibody assay miniaturizationto address the practical complexities of large-scalebiomarker screening of clinical cohorts By adaptingthe microarray printer to accommodate glycolipidswe were able to print arrays in a high throughputmanner One maximum capacity assay is capable ofprinting 320 individual arrays on a total of 20slides Conducting an equivalent screen on thissample size and antigen repertoire in a conventionalELISA study would require 640 ELISA plates (96-well) Miniaturization of the glycolipid assay allowsus to use 100-fold smaller volumes of patient serumthus the total serum use by microarray is 2 mL toprobe against 78 target antigens with duplicatemeasurement compared with 200 mL of serum fora comparable study performed on ELISA In additionthe use of fluorescently conjugated antibodies allowsfor multiplex screening whereby IgG andimmunoglobulin M can be detected simultaneouslyon each assay platform a benefit not routinelyavailable in ELISA readers
In this study a single electrophysiologic assess-ment was performed on 192 of the 266 patientsin the GBS cohort Of these patients 531 of pa-tients were classified as having an axonal variant ofGBS 302 of patients with AIDP while the
clinical variant of the remaining 167 of patientswas unclassified These diagnostic categorizationfrequencies are in line with a previously reportedGBS cohort from Bangladesh7 As found here thisprevious report also identified a low incidence ofpatient serum containing anti-GD1a antibodies(14) which was surprising given the high fre-quency of axonal GBS in comparison with studiesfrom elsewhere68ndash10 However probing Bangladeshsera against heteromeric complexes provedadvantageous as it revealed populations of complex-dependent GA1GD1a and GM1GD1a antibodieswhich were distinct from antibodies binding GD1a asa single antigen Similarly GM1GQ1b complex-dependent antibody binding was identified as a dis-tinct population in 119 of AMAN samples whencompared to GM1 and GQ1b alone Using a refinedpanel of only 3 antiglycolipid targets a high sensitivityand specificity was achieved for patients with AMANin this cohort
In addition to screening GBS patient samples 579combined control sera were examined This screenidentified variations in the baseline levels of antigan-glioside antibodies reactivities for each of the targetsWhen performing conventional ELISA the currentgold standard method for antiganglioside antibodydetection a universal optical density threshold of pos-itivity across all glycolipids is widely used In thisgeographically restricted study we can see that inorder to optimize both sensitivity and specificity ofthe assay individual threshold must be calculatedin order to determine the normal range of these nat-urally occurring and GBS-independent anticarbohy-drate antibodies
In this study we present data that identify thepresence of both single and heteromeric glycolipidcomplex binding antibodies in this large GBS cohortInitial biomarker screening employed a panel of 78targets however the selective inclusion of a very lim-ited panel of targets enabled optimization of both sen-sitivity and specificity due to significant targetredundancy resulting from the presence of polyclonalantibodies or cross-reactivity of specific antibodyspecies When considering the overwhelming com-plexities of identifying diagnostically important glyco-lipid complexes by ELISA screening the necessitiesof miniaturization are clearly evident for the initialanalysis in order to find the markers that are mostinformative
AUTHOR CONTRIBUTIONSSK Halstead study concept and design experimental work principal
draft of manuscript G Kalna statistical analysis and interpretation
MB Islam acquisition of clinical data I Jahan acquisition of clinical
data QD Mohammad acquisition of clinical data BC Jacobs study
concept and design study supervision HP Endtz study concept and
design study supervision Z Islam study concept and design critical
8 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
revision of the manuscript for important intellectual content HJ Willison
study concept and design critical revision of the manuscript for important
intellectual content study supervision
STUDY FUNDINGThis study is funded by the Wellcome Trust and the Chandra Mehta
Foundation
DISCLOSURESK Halstead G Kalna MB Islam I Jahan and QD Mohammad
report no disclosures BC Jacobs received travel funding from Baxter
International is on the editorial board for Journal of the Peripheral Ner-
vous System and received research support from Grifols CSL-Behring
Netherlands Organization for Health Research and Development Eras-
mus MC Prinses Beatrix GBS-CIDP Foundation International Prinses
Beatrix Fonds and Prinses Beatrix Spierfonds HP Endtz received
research support from EC Horizon 2020 Foundation Merieux Bill amp
Melinda Gates Foundation and GBS-CIDP foundation Z Islam reports
no disclosures HJ Willison is on the editorial board for Nature Clinical
Practice Neurology Journal of Neuroimmunology Muscle and Nerve Jour-
nal of Peripheral Nervous System Clinical and Experimental Neuroimmu-
nology and Experimental Neurology is an associate editor for BMC
Neurology holds a patent for combinatorial glycoarray technology and
received research support from Ipsen Annexon Alexion and MRC Go
to Neurologyorgnn for full disclosure forms
Received April 4 2016 Accepted in final form August 5 2016
REFERENCES1 Rinaldi S Brennan KM Willison HJ Heteromeric glyco-
lipid complexes as modulators of autoantibody and lectin
binding Prog Lipid Res 20104987ndash95
2 Kaida K Morita D Kanzaki M et al Ganglioside com-
plexes as new target antigens in Guillain-Barre syndrome
Ann Neurol 200456567ndash571
3 Galban-Horcajo F Halstead SK McGonigal R
Willison HJ The application of glycosphingolipid arrays
to autoantibody detection in neuroimmunological disor-
ders Curr Opin Chem Biol 20141878ndash86
4 Asbury AK Cornblath DR Assessment of current diag-
nostic criteria for Guillain-Barre syndrome Ann Neurol
1990(27 suppl)S21ndashS24
5 Jones CM Athanasiou T Summary receiver operating
characteristic curve analysis techniques in the evaluation
of diagnostic tests Ann Thorac Surg 20057916ndash20
6 Ho TW Willison HJ Nachamkin I et al Anti-GD1a
antibody is associated with axonal but not demyelinating
forms of Guillain-Barre syndrome Ann Neurol 199945
168ndash173
7 Islam Z Jacobs BC van Belkum A et al Axonal var-
iant of Guillain-Barre syndrome associated with cam-
pylobacter infection in Bangladesh Neurology 2010
74581ndash587
8 Kim JK Bae JS Kim DS et al Prevalence of anti-
ganglioside antibodies and their clinical correlates with
Guillain-Barreacute syndrome in Korea a nationwide multicen-
ter study J Clin Neurol 20141094ndash100
9 Willison HJ Yuki N Peripheral neuropathies and anti-
glycolipid antibodies Brain 20021252591ndash2625
10 Yuki N Yamada M Sato S et al Association of IgG anti-
GD1a antibody with severe Guillain-Barre syndrome
Muscle Nerve 199316642ndash647
Neurology Neuroimmunology amp Neuroinflammation 9
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
DOI 101212NXI000000000000028420163 Neurol Neuroimmunol Neuroinflamm
Susan K Halstead Gabriela Kalna Mohammad B Islam et al complexes
Microarray screening of Guillain-Barreacute syndrome sera for antibodies to glycolipid
This information is current as of September 28 2016
ServicesUpdated Information amp
httpnnneurologyorgcontent36e284fullhtmlincluding high resolution figures can be found at
Supplementary Material
httpnnneurologyorgcontentsuppl2017022136e284DC2 httpnnneurologyorgcontentsuppl2016092836e284DC1
Supplementary material can be found at
References httpnnneurologyorgcontent36e284fullhtmlref-list-1
This article cites 9 articles 0 of which you can access for free at
Subspecialty Collections
httpnnneurologyorgcgicollectionperipheral_neuropathyPeripheral neuropathy
httpnnneurologyorgcgicollectionguillainbarre_syndromeGuillain-Barre syndrome
httpnnneurologyorgcgicollectionautoimmune_diseasesAutoimmune diseasesfollowing 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
2016 American Academy of Neurology All rights reserved Online ISSN 2332-7812Published since April 2014 it is an open-access online-only continuous publication journal Copyright copy
is an official journal of the American Academy of NeurologyNeurol Neuroimmunol Neuroinflamm
following preceding symptoms (either in isolation orin combination with other symptoms) diarrhea 130cases (489) respiratory infection 59 cases(222) fever 34 cases (128) chicken pox 7cases (26) vaccination 6 cases (23) other 21cases (79) No preceding symptoms were reportedin 32 cases (120) There were 35 deaths (132)and 58 patients (218) required ventilation at somepoint during treatment For the family and neuro-logic control demographic and preceding symptomdata see the e-Methods
Antiglycolipid antibody screen The mean intensity val-ues were calculated from 4 intensity measurementsper target for each serum sample These 78 targetsare visually represented as a heat map employingPearson cluster coefficient and displayed as 3 groupsGBS (n 5 266) FC (n 5 258) and NC (n 5 321)with antibody binding intensity portrayed in a rain-bow scale red being high binding and blue being lowbinding (figure 1A) As expected for a GBS popula-tion in which multiple targets are known to be pres-ent no single antigen binding pattern was dominantthroughout the cohort examined Instead 30 targetswere found with significantly increased (p 005)binding intensities in GBS as compared indepen-dently with both FC and NC groups These dataon 30 significant targets are also displayed solely forthe GBS group for simplicity (figure 1B) Twenty-eight of the targets in the GBS group wereheteromeric complexes and 2 (GA1 and GM1)were single glycolipids While these 30 targets arehighly significantly different between disease andcontrol groups and may be clinically useful whenmultiple tests are performed screening for oneantigen or antigen complex cannot be used todetermine a measure of clinical usefulness byreceiver operating characteristic (ROC) analysis dueto the low sensitivity of individual targets Indeed asexpected when subjected to ROC analysis no targetsreached area under the curve $075 (the minimallevel considered to be the gold standard fora clinically useful biomarker5) when compared withcombined (family and neurologic) control groups
Certain targets (containing SGPG cholesteroland GA1) returned higher than average signals acrossboth disease and control groups indicating that nor-mal ranges need to be determined for each antigenOptimized positivity threshold values were calculatedfor each target antigen defined as the 95th percentileof the combined controls All intensity values wereconverted to binary data (positivenegative) deter-mined by the optimal threshold calculated for eachtarget and all subsequent data are expressed in termsof sensitivity and specificity When considering all 78targets a sensitivity of 827 and a specificity of
371 was achieved indicating that 629 of thecombined control samples had antibody bindingintensity greater than the threshold for one or moreof the 78 targets despite only 5 of controls beingpositive for any one individual antigen target In orderto increase the specificity of the assay we establisheda refined panel of target antigens
Anti-GM1 -GA1 -GD1a and -GQ1b antibodies Focus-ing on 4 glycolipids that are already recognized to betargets in GBS and were prominent antigens in thisscreen several common cluster patterns comprisingGM1 GA1 GD1a and GQ1b alone or in complexcould be identified (figure 2A) The largest group ofsera contained anti-GM1 antibodies (n 5 92sensitivity 5 346) Many of these also boundto GM1 when in heteromeric complex howeverin some cases binding to GM1 was inhibitedwhen in heteromeric complex with anotherglycolipid Considering the samples in whichheteromeric complexes enhanced anti-GM1antibody binding GM1GD1a and GM1GQ1bcomplexes were prominent pairings many of thesesamples containing completely complex-dependentantibodies (ie without the complex the antibodywould not be identifiable as there was no reactivityabove threshold for each individual ganglioside)Thus of the 98 (368) GBS sera that boundGM1GD1a complex 25 were completelycomplex dependent Of the 98 (368) GBS serathat bound GM1GQ1b complex 27 werecompletely complex dependent (figure 2 EndashF)
Antibody binding events were infrequent in theAMAN population for GA1 (37102 363) andGD1a (14102 137) when considered as singleantigens compared with other GBS populations inwhich anti-GD1a-positive AMAN cases are moreprominent6 However the GA1GD1a complex wasa significant target in 36 (353) AMAN samples ofwhich 10 were completely complex dependent (figure2 BndashD) Similarly GM1GD1a was a significant tar-get in 57 (559) AMAN samples of which 14 werecomplex dependent In this clinical population bothGA1 and GM1 therefore appear to enhance anti-GD1a antibody detection when the 2 glycolipidsare in complex
Complex enhancement and complex inhibition In orderto illustrate the enhancing and inhibiting effect ofcomplexes exemplary data are presented for GA1GD1a (figure 2 BndashD) and GM1GQ1b (figure 2EndashG) Heterogeneous patterns are found as ex-pected Thus there are 2 populations of anti-GA1GD1a antibodies in GBS sera (1) those inwhich binding intensities to GA1 or GD1a areattenuated by GA1GD1a complex and (2) thosein which binding is greatly enhanced by the
Neurology Neuroimmunology amp Neuroinflammation 3
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Figure 1 Heat map of antibody binding patterns to glycolipid targets in Guillain-Barreacute syndrome (GBS) casesand controls
(A) Patient (GBS) and control sera (family controls [FC] and neurologic disease controls [NC]) were screened against a panelof 78 single and heteromeric glycolipid targets on glycolipid microarray Mean fluorescent intensity values were calculatedand graphically presented as a heat map The rainbow scale indicates the intensity of the antibody binding in which blue isweak and red is strong antibody binding Pearson hierarchical clustering was employed to group samples with similar anti-gen binding patterns (B) Heat map presentation of GBS samples for the 30 glycolipid targets identified as having signifi-cantly different binding intensities between GBS and FC or NC groups (p 005)
4 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Figure 2 Subanalysis of 10 major antiglycolipid antibody targets in Guillain-Barreacute syndrome (GBS) cases
(A) Heat map presentation of patient (GBS) and control sera (family controls [FC] and neurologic disease controls [NC]) reactivity with GM1 GD1a GQ1b andGA1 as single antigens and their 6 possible heteromeric complexes When selecting this small panel of 10 targets as serum biomarkers of GBS in this clinicalcohort the combined sensitivity is 643 and the specificity is 771 (B) Dot plot presentation of GBS (n5266) antibody binding intensities for single GA1and GD1a and their heteromeric complex GA1GD1a (C) Heat map presentation of patient binding intensities Each row represents a single patient withbinding intensity to each target represented using the rainbow scale (D) Line graph compares the binding intensity for complex GA1GD1a with the sum ofthe single antigens (GA1 and GD1a) Green lines indicates complex enhancement while red lines represent complex attenuation (E) Dot plot presentation ofGBS (n 5 266) antibody binding intensities for single GM1 and GQ1b and their heteromeric complex GM1GQ1b (F) Heat map presentation of patientbinding intensities Each row represents a single patient with binding to each target represented using the rainbow scale (G) Line graph compares thebinding intensity for complex GM1GQ1b with the sum of the single antigens (GM1 and GQ1b) Green lines indicates complex enhancement while red linesrepresent complex attenuation
Neurology Neuroimmunology amp Neuroinflammation 5
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
complex most usually in a completely complex-dependent fashion (ie no binding is present toeither partner alone figure 2D green lines)Similarly when considering GM1 and GQ1bwhile some anti-GM1 and anti-GQ1b sera haveattenuated binding intensities when in GM1GQ1b complex there is a significantly largegroup of sera in which major enhancement
occurs in the presence of the complex (figure 2Ggreen lines)
The added value and limitations of screening for
heteromeric complexes To assess the overall impactof including glycolipid heteromeric complexes inthe screening platform we compared the number ofGBS samples classified as positive (ie returning anintensity value 95th percentile of the combinedcontrols) for single glycolipids and heteromeric com-plexes for the 10 frequently observed reactivities (fig-ure 2A) A total of 131 of 266 GBS samples and 87 of579 combined controls were positive for one or moresingle targets (sensitivity 492 specificity 850)whereas 155266 GBS samples and 95579 com-bined controls were positive for one or more hetero-meric complexes (sensitivity 583 specificity836) Therefore by screening for heteromericcomplexes a gain of 91 in sensitivity (p 5
00021) is achieved without a significant loss in spec-ificity (14 p 5 0445) Screening of GBS serairrespective of the clinical variant (AMAN AIDPor unclassified) against this discrete panel of 10 gly-colipids (4 single and 6 complexes) resulted in anoverall assay sensitivity of 643 and specificity of771
However when examining a larger panel of tar-gets a gain in sensitivity is frequently offset by anequivalent or greater loss in specificity (figure 3A)For example when examining the 30 targets withsignificantly different fluorescent intensity values (fig-ure 1B) a gain of 304 in sensitivity (from 410to 714) is offset by a loss of 29 specificity (from912 to 622) when comparing single and com-plex positivity frequency Similarly when consideringall 78 targets printed on the glycolipid arrays there isa 147 gain in sensitivity (from 650 to 797)while specificity drops 243 (from 658 to415) for single and complexes respectively Thesedata show that when examining the data in its total-ity the disease specificity of a very broad unbiasedscreen is expectedly poor owing to the large numberof controls whose sera contain at least one species ofantiglycolipid antibody above the normal range(95th percentile)
Antiglycolipid antibodies associated with clinical
variants Correlation between antiglycolipid anti-body profiles and clinical features are presented intable and as e-Results Having determined a panelof 10 glycolipid reactivities optimized for sensitivityand specificity for all patients with GBS irrespectiveof the clinical variant next we sought to determinewhether the presence of specific antibody reactiv-ities would segregate with a clinical variant or couldbe used to predict the clinical phenotype or out-come (see e-Results) Patients who had unclassified
Figure 3 Sensitivity and specificity of microarray analysis in relation to targetnumber
(A) Comparison of overall assay accuracy (both sensitivity and specificity) when screeningagainst an increasing number of target antigens Assay specificity declines as the numbersof target antigens increases Improved assay performance is achieved when a small panel of10 targets is selected (B) Dot plot compares the maximum binding intensity signal for eachpatient or control when comparing single and heteromeric complexes GBS patient medianarbitrary binding intensity for single targets (7473 FIU) is significantly less than for hetero-meric complex targets (35899 FIU p 00001) indicating that heteromeric complexesreturn higher signals than single antigens
6 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
GBS variants (n 5 105) were excluded from thisanalysis Six targets were identified as being signif-icant in the AIDP population when compared withthe AMAN clinical variant These included SGPGalone as well as 2 complexes containing SGPG(SGPGGQ1b and SGPGChol) In addition 2heteromeric complexes containing LM1 (LM1Sul-phatide and LM1Chol) and CTHChol were sig-nificantly associated with AIDP (p 005) Whilesignificant each of these targets was present at low
frequencies within the patient samples and thusindividually they had low diagnostic sensitivity(ranging from 105 to 140) but with a highspecificity both for combined controls (950) andpatients with AMAN (ranging from 960 to980 depending upon the target) When com-bined as a diagnostic panel of 6 antigen targets forAIDP a sensitivity of 339 and a specificity of873 and 834 (for AMAN and controls respec-tively) was reached
Table Glycolipid reactivities associated with acute inflammatory demyelinating polyneuropathy (AIDP)andacute motor axonal neuropathy (AMAN) clinical variants
AIDP (n 5 59) AMAN (n 5 102)
Guillain-Barreacute syndrome variant biomarker SGPG GM1 single 1 All GM1 complexes
SGPGGQ1b GA1 single 1 All GA1 complexes (excl GA1GQ1b)
SGPGChol GD1bGD1a
LM1Sulph GD1bPS
LM1Chol GD1bChol
CTHChol GD1bSulph
Ocular deficits GQ1bGD1b NA
GQ1bSGPG
GQ1bCTH
GQ1bChol
GQ1bGalC
LM1CTH
Bulbar palsy NA GQ1bPS
Facial palsy NA GA1GD1a
Figure 4 Heat map of antibody binding patterns in axonal vs demyelinating Guillain-Barreacute syndrome (GBS)cases
(A) Heat map presentation of the 26 glycolipid targets significant for the acute motor axonal neuropathy variant of GBScompared with the demyelinating (acute inflammatory demyelinating polyneuropathy) clinical subtype (B) Refining thescreening targets to a panel of 3 heteromeric complexes (GM1GQ1b GM1phosphatidylserine and GA1GD1a) resultsin a sensitivity of 784 and a specificity of 877 (for combined controls)
Neurology Neuroimmunology amp Neuroinflammation 7
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Twenty-six targets were identified as being sig-nificant (p 005) in AMAN when compared withthe AIDP population (figure 4A) These includeGM1 alone plus GM1 in complex with all testedlipids (GM1GA1 GM1GD1a GM1GD1bGM1GQ1b GM1SGPG GM1LM1 GM1CTH GM1PS GM1Chol GM1SulphatideGM1GalC) GA1 alone plus GA1 in complex withall tested lipids with the exception of GA1GQ1b(GA1GD1a GA1GD1b GA1SGPG GA1LM1GA1CTH GA1PS GA1Chol GA1SulphatideGA1GalC) GD1bGD1a GD1bPS GD1bCholand GD1bSulphatide Of these GM1GD1a hadthe highest sensitivity (559) and a high specific-ity for both combined controls (950) and pa-tients with AIDP (848) When combined asa diagnostic panel of 26 antigen targets for AMANa sensitivity of 863 and a specificity of 644and 458 (for combined controls and AIDPrespectively) was reached As previously demon-strated the specificity was reduced with increasingantigen targets therefore we sought to refine thenumber of antigen targets in order to optimize sen-sitivity and specificity When selecting only 3 anti-gen targets (GM1GQ1b GM1PS and GA1GD1a) a sensitivity of 784 and specificity of877 and 610 (for combined control andAIDP respectively) was achieved (figure 4B)
DISCUSSION This study applies technical develop-ments in antiglycolipid antibody assay miniaturizationto address the practical complexities of large-scalebiomarker screening of clinical cohorts By adaptingthe microarray printer to accommodate glycolipidswe were able to print arrays in a high throughputmanner One maximum capacity assay is capable ofprinting 320 individual arrays on a total of 20slides Conducting an equivalent screen on thissample size and antigen repertoire in a conventionalELISA study would require 640 ELISA plates (96-well) Miniaturization of the glycolipid assay allowsus to use 100-fold smaller volumes of patient serumthus the total serum use by microarray is 2 mL toprobe against 78 target antigens with duplicatemeasurement compared with 200 mL of serum fora comparable study performed on ELISA In additionthe use of fluorescently conjugated antibodies allowsfor multiplex screening whereby IgG andimmunoglobulin M can be detected simultaneouslyon each assay platform a benefit not routinelyavailable in ELISA readers
In this study a single electrophysiologic assess-ment was performed on 192 of the 266 patientsin the GBS cohort Of these patients 531 of pa-tients were classified as having an axonal variant ofGBS 302 of patients with AIDP while the
clinical variant of the remaining 167 of patientswas unclassified These diagnostic categorizationfrequencies are in line with a previously reportedGBS cohort from Bangladesh7 As found here thisprevious report also identified a low incidence ofpatient serum containing anti-GD1a antibodies(14) which was surprising given the high fre-quency of axonal GBS in comparison with studiesfrom elsewhere68ndash10 However probing Bangladeshsera against heteromeric complexes provedadvantageous as it revealed populations of complex-dependent GA1GD1a and GM1GD1a antibodieswhich were distinct from antibodies binding GD1a asa single antigen Similarly GM1GQ1b complex-dependent antibody binding was identified as a dis-tinct population in 119 of AMAN samples whencompared to GM1 and GQ1b alone Using a refinedpanel of only 3 antiglycolipid targets a high sensitivityand specificity was achieved for patients with AMANin this cohort
In addition to screening GBS patient samples 579combined control sera were examined This screenidentified variations in the baseline levels of antigan-glioside antibodies reactivities for each of the targetsWhen performing conventional ELISA the currentgold standard method for antiganglioside antibodydetection a universal optical density threshold of pos-itivity across all glycolipids is widely used In thisgeographically restricted study we can see that inorder to optimize both sensitivity and specificity ofthe assay individual threshold must be calculatedin order to determine the normal range of these nat-urally occurring and GBS-independent anticarbohy-drate antibodies
In this study we present data that identify thepresence of both single and heteromeric glycolipidcomplex binding antibodies in this large GBS cohortInitial biomarker screening employed a panel of 78targets however the selective inclusion of a very lim-ited panel of targets enabled optimization of both sen-sitivity and specificity due to significant targetredundancy resulting from the presence of polyclonalantibodies or cross-reactivity of specific antibodyspecies When considering the overwhelming com-plexities of identifying diagnostically important glyco-lipid complexes by ELISA screening the necessitiesof miniaturization are clearly evident for the initialanalysis in order to find the markers that are mostinformative
AUTHOR CONTRIBUTIONSSK Halstead study concept and design experimental work principal
draft of manuscript G Kalna statistical analysis and interpretation
MB Islam acquisition of clinical data I Jahan acquisition of clinical
data QD Mohammad acquisition of clinical data BC Jacobs study
concept and design study supervision HP Endtz study concept and
design study supervision Z Islam study concept and design critical
8 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
revision of the manuscript for important intellectual content HJ Willison
study concept and design critical revision of the manuscript for important
intellectual content study supervision
STUDY FUNDINGThis study is funded by the Wellcome Trust and the Chandra Mehta
Foundation
DISCLOSURESK Halstead G Kalna MB Islam I Jahan and QD Mohammad
report no disclosures BC Jacobs received travel funding from Baxter
International is on the editorial board for Journal of the Peripheral Ner-
vous System and received research support from Grifols CSL-Behring
Netherlands Organization for Health Research and Development Eras-
mus MC Prinses Beatrix GBS-CIDP Foundation International Prinses
Beatrix Fonds and Prinses Beatrix Spierfonds HP Endtz received
research support from EC Horizon 2020 Foundation Merieux Bill amp
Melinda Gates Foundation and GBS-CIDP foundation Z Islam reports
no disclosures HJ Willison is on the editorial board for Nature Clinical
Practice Neurology Journal of Neuroimmunology Muscle and Nerve Jour-
nal of Peripheral Nervous System Clinical and Experimental Neuroimmu-
nology and Experimental Neurology is an associate editor for BMC
Neurology holds a patent for combinatorial glycoarray technology and
received research support from Ipsen Annexon Alexion and MRC Go
to Neurologyorgnn for full disclosure forms
Received April 4 2016 Accepted in final form August 5 2016
REFERENCES1 Rinaldi S Brennan KM Willison HJ Heteromeric glyco-
lipid complexes as modulators of autoantibody and lectin
binding Prog Lipid Res 20104987ndash95
2 Kaida K Morita D Kanzaki M et al Ganglioside com-
plexes as new target antigens in Guillain-Barre syndrome
Ann Neurol 200456567ndash571
3 Galban-Horcajo F Halstead SK McGonigal R
Willison HJ The application of glycosphingolipid arrays
to autoantibody detection in neuroimmunological disor-
ders Curr Opin Chem Biol 20141878ndash86
4 Asbury AK Cornblath DR Assessment of current diag-
nostic criteria for Guillain-Barre syndrome Ann Neurol
1990(27 suppl)S21ndashS24
5 Jones CM Athanasiou T Summary receiver operating
characteristic curve analysis techniques in the evaluation
of diagnostic tests Ann Thorac Surg 20057916ndash20
6 Ho TW Willison HJ Nachamkin I et al Anti-GD1a
antibody is associated with axonal but not demyelinating
forms of Guillain-Barre syndrome Ann Neurol 199945
168ndash173
7 Islam Z Jacobs BC van Belkum A et al Axonal var-
iant of Guillain-Barre syndrome associated with cam-
pylobacter infection in Bangladesh Neurology 2010
74581ndash587
8 Kim JK Bae JS Kim DS et al Prevalence of anti-
ganglioside antibodies and their clinical correlates with
Guillain-Barreacute syndrome in Korea a nationwide multicen-
ter study J Clin Neurol 20141094ndash100
9 Willison HJ Yuki N Peripheral neuropathies and anti-
glycolipid antibodies Brain 20021252591ndash2625
10 Yuki N Yamada M Sato S et al Association of IgG anti-
GD1a antibody with severe Guillain-Barre syndrome
Muscle Nerve 199316642ndash647
Neurology Neuroimmunology amp Neuroinflammation 9
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
DOI 101212NXI000000000000028420163 Neurol Neuroimmunol Neuroinflamm
Susan K Halstead Gabriela Kalna Mohammad B Islam et al complexes
Microarray screening of Guillain-Barreacute syndrome sera for antibodies to glycolipid
This information is current as of September 28 2016
ServicesUpdated Information amp
httpnnneurologyorgcontent36e284fullhtmlincluding high resolution figures can be found at
Supplementary Material
httpnnneurologyorgcontentsuppl2017022136e284DC2 httpnnneurologyorgcontentsuppl2016092836e284DC1
Supplementary material can be found at
References httpnnneurologyorgcontent36e284fullhtmlref-list-1
This article cites 9 articles 0 of which you can access for free at
Subspecialty Collections
httpnnneurologyorgcgicollectionperipheral_neuropathyPeripheral neuropathy
httpnnneurologyorgcgicollectionguillainbarre_syndromeGuillain-Barre syndrome
httpnnneurologyorgcgicollectionautoimmune_diseasesAutoimmune diseasesfollowing 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
2016 American Academy of Neurology All rights reserved Online ISSN 2332-7812Published since April 2014 it is an open-access online-only continuous publication journal Copyright copy
is an official journal of the American Academy of NeurologyNeurol Neuroimmunol Neuroinflamm
Figure 1 Heat map of antibody binding patterns to glycolipid targets in Guillain-Barreacute syndrome (GBS) casesand controls
(A) Patient (GBS) and control sera (family controls [FC] and neurologic disease controls [NC]) were screened against a panelof 78 single and heteromeric glycolipid targets on glycolipid microarray Mean fluorescent intensity values were calculatedand graphically presented as a heat map The rainbow scale indicates the intensity of the antibody binding in which blue isweak and red is strong antibody binding Pearson hierarchical clustering was employed to group samples with similar anti-gen binding patterns (B) Heat map presentation of GBS samples for the 30 glycolipid targets identified as having signifi-cantly different binding intensities between GBS and FC or NC groups (p 005)
4 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Figure 2 Subanalysis of 10 major antiglycolipid antibody targets in Guillain-Barreacute syndrome (GBS) cases
(A) Heat map presentation of patient (GBS) and control sera (family controls [FC] and neurologic disease controls [NC]) reactivity with GM1 GD1a GQ1b andGA1 as single antigens and their 6 possible heteromeric complexes When selecting this small panel of 10 targets as serum biomarkers of GBS in this clinicalcohort the combined sensitivity is 643 and the specificity is 771 (B) Dot plot presentation of GBS (n5266) antibody binding intensities for single GA1and GD1a and their heteromeric complex GA1GD1a (C) Heat map presentation of patient binding intensities Each row represents a single patient withbinding intensity to each target represented using the rainbow scale (D) Line graph compares the binding intensity for complex GA1GD1a with the sum ofthe single antigens (GA1 and GD1a) Green lines indicates complex enhancement while red lines represent complex attenuation (E) Dot plot presentation ofGBS (n 5 266) antibody binding intensities for single GM1 and GQ1b and their heteromeric complex GM1GQ1b (F) Heat map presentation of patientbinding intensities Each row represents a single patient with binding to each target represented using the rainbow scale (G) Line graph compares thebinding intensity for complex GM1GQ1b with the sum of the single antigens (GM1 and GQ1b) Green lines indicates complex enhancement while red linesrepresent complex attenuation
Neurology Neuroimmunology amp Neuroinflammation 5
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
complex most usually in a completely complex-dependent fashion (ie no binding is present toeither partner alone figure 2D green lines)Similarly when considering GM1 and GQ1bwhile some anti-GM1 and anti-GQ1b sera haveattenuated binding intensities when in GM1GQ1b complex there is a significantly largegroup of sera in which major enhancement
occurs in the presence of the complex (figure 2Ggreen lines)
The added value and limitations of screening for
heteromeric complexes To assess the overall impactof including glycolipid heteromeric complexes inthe screening platform we compared the number ofGBS samples classified as positive (ie returning anintensity value 95th percentile of the combinedcontrols) for single glycolipids and heteromeric com-plexes for the 10 frequently observed reactivities (fig-ure 2A) A total of 131 of 266 GBS samples and 87 of579 combined controls were positive for one or moresingle targets (sensitivity 492 specificity 850)whereas 155266 GBS samples and 95579 com-bined controls were positive for one or more hetero-meric complexes (sensitivity 583 specificity836) Therefore by screening for heteromericcomplexes a gain of 91 in sensitivity (p 5
00021) is achieved without a significant loss in spec-ificity (14 p 5 0445) Screening of GBS serairrespective of the clinical variant (AMAN AIDPor unclassified) against this discrete panel of 10 gly-colipids (4 single and 6 complexes) resulted in anoverall assay sensitivity of 643 and specificity of771
However when examining a larger panel of tar-gets a gain in sensitivity is frequently offset by anequivalent or greater loss in specificity (figure 3A)For example when examining the 30 targets withsignificantly different fluorescent intensity values (fig-ure 1B) a gain of 304 in sensitivity (from 410to 714) is offset by a loss of 29 specificity (from912 to 622) when comparing single and com-plex positivity frequency Similarly when consideringall 78 targets printed on the glycolipid arrays there isa 147 gain in sensitivity (from 650 to 797)while specificity drops 243 (from 658 to415) for single and complexes respectively Thesedata show that when examining the data in its total-ity the disease specificity of a very broad unbiasedscreen is expectedly poor owing to the large numberof controls whose sera contain at least one species ofantiglycolipid antibody above the normal range(95th percentile)
Antiglycolipid antibodies associated with clinical
variants Correlation between antiglycolipid anti-body profiles and clinical features are presented intable and as e-Results Having determined a panelof 10 glycolipid reactivities optimized for sensitivityand specificity for all patients with GBS irrespectiveof the clinical variant next we sought to determinewhether the presence of specific antibody reactiv-ities would segregate with a clinical variant or couldbe used to predict the clinical phenotype or out-come (see e-Results) Patients who had unclassified
Figure 3 Sensitivity and specificity of microarray analysis in relation to targetnumber
(A) Comparison of overall assay accuracy (both sensitivity and specificity) when screeningagainst an increasing number of target antigens Assay specificity declines as the numbersof target antigens increases Improved assay performance is achieved when a small panel of10 targets is selected (B) Dot plot compares the maximum binding intensity signal for eachpatient or control when comparing single and heteromeric complexes GBS patient medianarbitrary binding intensity for single targets (7473 FIU) is significantly less than for hetero-meric complex targets (35899 FIU p 00001) indicating that heteromeric complexesreturn higher signals than single antigens
6 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
GBS variants (n 5 105) were excluded from thisanalysis Six targets were identified as being signif-icant in the AIDP population when compared withthe AMAN clinical variant These included SGPGalone as well as 2 complexes containing SGPG(SGPGGQ1b and SGPGChol) In addition 2heteromeric complexes containing LM1 (LM1Sul-phatide and LM1Chol) and CTHChol were sig-nificantly associated with AIDP (p 005) Whilesignificant each of these targets was present at low
frequencies within the patient samples and thusindividually they had low diagnostic sensitivity(ranging from 105 to 140) but with a highspecificity both for combined controls (950) andpatients with AMAN (ranging from 960 to980 depending upon the target) When com-bined as a diagnostic panel of 6 antigen targets forAIDP a sensitivity of 339 and a specificity of873 and 834 (for AMAN and controls respec-tively) was reached
Table Glycolipid reactivities associated with acute inflammatory demyelinating polyneuropathy (AIDP)andacute motor axonal neuropathy (AMAN) clinical variants
AIDP (n 5 59) AMAN (n 5 102)
Guillain-Barreacute syndrome variant biomarker SGPG GM1 single 1 All GM1 complexes
SGPGGQ1b GA1 single 1 All GA1 complexes (excl GA1GQ1b)
SGPGChol GD1bGD1a
LM1Sulph GD1bPS
LM1Chol GD1bChol
CTHChol GD1bSulph
Ocular deficits GQ1bGD1b NA
GQ1bSGPG
GQ1bCTH
GQ1bChol
GQ1bGalC
LM1CTH
Bulbar palsy NA GQ1bPS
Facial palsy NA GA1GD1a
Figure 4 Heat map of antibody binding patterns in axonal vs demyelinating Guillain-Barreacute syndrome (GBS)cases
(A) Heat map presentation of the 26 glycolipid targets significant for the acute motor axonal neuropathy variant of GBScompared with the demyelinating (acute inflammatory demyelinating polyneuropathy) clinical subtype (B) Refining thescreening targets to a panel of 3 heteromeric complexes (GM1GQ1b GM1phosphatidylserine and GA1GD1a) resultsin a sensitivity of 784 and a specificity of 877 (for combined controls)
Neurology Neuroimmunology amp Neuroinflammation 7
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Twenty-six targets were identified as being sig-nificant (p 005) in AMAN when compared withthe AIDP population (figure 4A) These includeGM1 alone plus GM1 in complex with all testedlipids (GM1GA1 GM1GD1a GM1GD1bGM1GQ1b GM1SGPG GM1LM1 GM1CTH GM1PS GM1Chol GM1SulphatideGM1GalC) GA1 alone plus GA1 in complex withall tested lipids with the exception of GA1GQ1b(GA1GD1a GA1GD1b GA1SGPG GA1LM1GA1CTH GA1PS GA1Chol GA1SulphatideGA1GalC) GD1bGD1a GD1bPS GD1bCholand GD1bSulphatide Of these GM1GD1a hadthe highest sensitivity (559) and a high specific-ity for both combined controls (950) and pa-tients with AIDP (848) When combined asa diagnostic panel of 26 antigen targets for AMANa sensitivity of 863 and a specificity of 644and 458 (for combined controls and AIDPrespectively) was reached As previously demon-strated the specificity was reduced with increasingantigen targets therefore we sought to refine thenumber of antigen targets in order to optimize sen-sitivity and specificity When selecting only 3 anti-gen targets (GM1GQ1b GM1PS and GA1GD1a) a sensitivity of 784 and specificity of877 and 610 (for combined control andAIDP respectively) was achieved (figure 4B)
DISCUSSION This study applies technical develop-ments in antiglycolipid antibody assay miniaturizationto address the practical complexities of large-scalebiomarker screening of clinical cohorts By adaptingthe microarray printer to accommodate glycolipidswe were able to print arrays in a high throughputmanner One maximum capacity assay is capable ofprinting 320 individual arrays on a total of 20slides Conducting an equivalent screen on thissample size and antigen repertoire in a conventionalELISA study would require 640 ELISA plates (96-well) Miniaturization of the glycolipid assay allowsus to use 100-fold smaller volumes of patient serumthus the total serum use by microarray is 2 mL toprobe against 78 target antigens with duplicatemeasurement compared with 200 mL of serum fora comparable study performed on ELISA In additionthe use of fluorescently conjugated antibodies allowsfor multiplex screening whereby IgG andimmunoglobulin M can be detected simultaneouslyon each assay platform a benefit not routinelyavailable in ELISA readers
In this study a single electrophysiologic assess-ment was performed on 192 of the 266 patientsin the GBS cohort Of these patients 531 of pa-tients were classified as having an axonal variant ofGBS 302 of patients with AIDP while the
clinical variant of the remaining 167 of patientswas unclassified These diagnostic categorizationfrequencies are in line with a previously reportedGBS cohort from Bangladesh7 As found here thisprevious report also identified a low incidence ofpatient serum containing anti-GD1a antibodies(14) which was surprising given the high fre-quency of axonal GBS in comparison with studiesfrom elsewhere68ndash10 However probing Bangladeshsera against heteromeric complexes provedadvantageous as it revealed populations of complex-dependent GA1GD1a and GM1GD1a antibodieswhich were distinct from antibodies binding GD1a asa single antigen Similarly GM1GQ1b complex-dependent antibody binding was identified as a dis-tinct population in 119 of AMAN samples whencompared to GM1 and GQ1b alone Using a refinedpanel of only 3 antiglycolipid targets a high sensitivityand specificity was achieved for patients with AMANin this cohort
In addition to screening GBS patient samples 579combined control sera were examined This screenidentified variations in the baseline levels of antigan-glioside antibodies reactivities for each of the targetsWhen performing conventional ELISA the currentgold standard method for antiganglioside antibodydetection a universal optical density threshold of pos-itivity across all glycolipids is widely used In thisgeographically restricted study we can see that inorder to optimize both sensitivity and specificity ofthe assay individual threshold must be calculatedin order to determine the normal range of these nat-urally occurring and GBS-independent anticarbohy-drate antibodies
In this study we present data that identify thepresence of both single and heteromeric glycolipidcomplex binding antibodies in this large GBS cohortInitial biomarker screening employed a panel of 78targets however the selective inclusion of a very lim-ited panel of targets enabled optimization of both sen-sitivity and specificity due to significant targetredundancy resulting from the presence of polyclonalantibodies or cross-reactivity of specific antibodyspecies When considering the overwhelming com-plexities of identifying diagnostically important glyco-lipid complexes by ELISA screening the necessitiesof miniaturization are clearly evident for the initialanalysis in order to find the markers that are mostinformative
AUTHOR CONTRIBUTIONSSK Halstead study concept and design experimental work principal
draft of manuscript G Kalna statistical analysis and interpretation
MB Islam acquisition of clinical data I Jahan acquisition of clinical
data QD Mohammad acquisition of clinical data BC Jacobs study
concept and design study supervision HP Endtz study concept and
design study supervision Z Islam study concept and design critical
8 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
revision of the manuscript for important intellectual content HJ Willison
study concept and design critical revision of the manuscript for important
intellectual content study supervision
STUDY FUNDINGThis study is funded by the Wellcome Trust and the Chandra Mehta
Foundation
DISCLOSURESK Halstead G Kalna MB Islam I Jahan and QD Mohammad
report no disclosures BC Jacobs received travel funding from Baxter
International is on the editorial board for Journal of the Peripheral Ner-
vous System and received research support from Grifols CSL-Behring
Netherlands Organization for Health Research and Development Eras-
mus MC Prinses Beatrix GBS-CIDP Foundation International Prinses
Beatrix Fonds and Prinses Beatrix Spierfonds HP Endtz received
research support from EC Horizon 2020 Foundation Merieux Bill amp
Melinda Gates Foundation and GBS-CIDP foundation Z Islam reports
no disclosures HJ Willison is on the editorial board for Nature Clinical
Practice Neurology Journal of Neuroimmunology Muscle and Nerve Jour-
nal of Peripheral Nervous System Clinical and Experimental Neuroimmu-
nology and Experimental Neurology is an associate editor for BMC
Neurology holds a patent for combinatorial glycoarray technology and
received research support from Ipsen Annexon Alexion and MRC Go
to Neurologyorgnn for full disclosure forms
Received April 4 2016 Accepted in final form August 5 2016
REFERENCES1 Rinaldi S Brennan KM Willison HJ Heteromeric glyco-
lipid complexes as modulators of autoantibody and lectin
binding Prog Lipid Res 20104987ndash95
2 Kaida K Morita D Kanzaki M et al Ganglioside com-
plexes as new target antigens in Guillain-Barre syndrome
Ann Neurol 200456567ndash571
3 Galban-Horcajo F Halstead SK McGonigal R
Willison HJ The application of glycosphingolipid arrays
to autoantibody detection in neuroimmunological disor-
ders Curr Opin Chem Biol 20141878ndash86
4 Asbury AK Cornblath DR Assessment of current diag-
nostic criteria for Guillain-Barre syndrome Ann Neurol
1990(27 suppl)S21ndashS24
5 Jones CM Athanasiou T Summary receiver operating
characteristic curve analysis techniques in the evaluation
of diagnostic tests Ann Thorac Surg 20057916ndash20
6 Ho TW Willison HJ Nachamkin I et al Anti-GD1a
antibody is associated with axonal but not demyelinating
forms of Guillain-Barre syndrome Ann Neurol 199945
168ndash173
7 Islam Z Jacobs BC van Belkum A et al Axonal var-
iant of Guillain-Barre syndrome associated with cam-
pylobacter infection in Bangladesh Neurology 2010
74581ndash587
8 Kim JK Bae JS Kim DS et al Prevalence of anti-
ganglioside antibodies and their clinical correlates with
Guillain-Barreacute syndrome in Korea a nationwide multicen-
ter study J Clin Neurol 20141094ndash100
9 Willison HJ Yuki N Peripheral neuropathies and anti-
glycolipid antibodies Brain 20021252591ndash2625
10 Yuki N Yamada M Sato S et al Association of IgG anti-
GD1a antibody with severe Guillain-Barre syndrome
Muscle Nerve 199316642ndash647
Neurology Neuroimmunology amp Neuroinflammation 9
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
DOI 101212NXI000000000000028420163 Neurol Neuroimmunol Neuroinflamm
Susan K Halstead Gabriela Kalna Mohammad B Islam et al complexes
Microarray screening of Guillain-Barreacute syndrome sera for antibodies to glycolipid
This information is current as of September 28 2016
ServicesUpdated Information amp
httpnnneurologyorgcontent36e284fullhtmlincluding high resolution figures can be found at
Supplementary Material
httpnnneurologyorgcontentsuppl2017022136e284DC2 httpnnneurologyorgcontentsuppl2016092836e284DC1
Supplementary material can be found at
References httpnnneurologyorgcontent36e284fullhtmlref-list-1
This article cites 9 articles 0 of which you can access for free at
Subspecialty Collections
httpnnneurologyorgcgicollectionperipheral_neuropathyPeripheral neuropathy
httpnnneurologyorgcgicollectionguillainbarre_syndromeGuillain-Barre syndrome
httpnnneurologyorgcgicollectionautoimmune_diseasesAutoimmune diseasesfollowing 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
2016 American Academy of Neurology All rights reserved Online ISSN 2332-7812Published since April 2014 it is an open-access online-only continuous publication journal Copyright copy
is an official journal of the American Academy of NeurologyNeurol Neuroimmunol Neuroinflamm
Figure 2 Subanalysis of 10 major antiglycolipid antibody targets in Guillain-Barreacute syndrome (GBS) cases
(A) Heat map presentation of patient (GBS) and control sera (family controls [FC] and neurologic disease controls [NC]) reactivity with GM1 GD1a GQ1b andGA1 as single antigens and their 6 possible heteromeric complexes When selecting this small panel of 10 targets as serum biomarkers of GBS in this clinicalcohort the combined sensitivity is 643 and the specificity is 771 (B) Dot plot presentation of GBS (n5266) antibody binding intensities for single GA1and GD1a and their heteromeric complex GA1GD1a (C) Heat map presentation of patient binding intensities Each row represents a single patient withbinding intensity to each target represented using the rainbow scale (D) Line graph compares the binding intensity for complex GA1GD1a with the sum ofthe single antigens (GA1 and GD1a) Green lines indicates complex enhancement while red lines represent complex attenuation (E) Dot plot presentation ofGBS (n 5 266) antibody binding intensities for single GM1 and GQ1b and their heteromeric complex GM1GQ1b (F) Heat map presentation of patientbinding intensities Each row represents a single patient with binding to each target represented using the rainbow scale (G) Line graph compares thebinding intensity for complex GM1GQ1b with the sum of the single antigens (GM1 and GQ1b) Green lines indicates complex enhancement while red linesrepresent complex attenuation
Neurology Neuroimmunology amp Neuroinflammation 5
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
complex most usually in a completely complex-dependent fashion (ie no binding is present toeither partner alone figure 2D green lines)Similarly when considering GM1 and GQ1bwhile some anti-GM1 and anti-GQ1b sera haveattenuated binding intensities when in GM1GQ1b complex there is a significantly largegroup of sera in which major enhancement
occurs in the presence of the complex (figure 2Ggreen lines)
The added value and limitations of screening for
heteromeric complexes To assess the overall impactof including glycolipid heteromeric complexes inthe screening platform we compared the number ofGBS samples classified as positive (ie returning anintensity value 95th percentile of the combinedcontrols) for single glycolipids and heteromeric com-plexes for the 10 frequently observed reactivities (fig-ure 2A) A total of 131 of 266 GBS samples and 87 of579 combined controls were positive for one or moresingle targets (sensitivity 492 specificity 850)whereas 155266 GBS samples and 95579 com-bined controls were positive for one or more hetero-meric complexes (sensitivity 583 specificity836) Therefore by screening for heteromericcomplexes a gain of 91 in sensitivity (p 5
00021) is achieved without a significant loss in spec-ificity (14 p 5 0445) Screening of GBS serairrespective of the clinical variant (AMAN AIDPor unclassified) against this discrete panel of 10 gly-colipids (4 single and 6 complexes) resulted in anoverall assay sensitivity of 643 and specificity of771
However when examining a larger panel of tar-gets a gain in sensitivity is frequently offset by anequivalent or greater loss in specificity (figure 3A)For example when examining the 30 targets withsignificantly different fluorescent intensity values (fig-ure 1B) a gain of 304 in sensitivity (from 410to 714) is offset by a loss of 29 specificity (from912 to 622) when comparing single and com-plex positivity frequency Similarly when consideringall 78 targets printed on the glycolipid arrays there isa 147 gain in sensitivity (from 650 to 797)while specificity drops 243 (from 658 to415) for single and complexes respectively Thesedata show that when examining the data in its total-ity the disease specificity of a very broad unbiasedscreen is expectedly poor owing to the large numberof controls whose sera contain at least one species ofantiglycolipid antibody above the normal range(95th percentile)
Antiglycolipid antibodies associated with clinical
variants Correlation between antiglycolipid anti-body profiles and clinical features are presented intable and as e-Results Having determined a panelof 10 glycolipid reactivities optimized for sensitivityand specificity for all patients with GBS irrespectiveof the clinical variant next we sought to determinewhether the presence of specific antibody reactiv-ities would segregate with a clinical variant or couldbe used to predict the clinical phenotype or out-come (see e-Results) Patients who had unclassified
Figure 3 Sensitivity and specificity of microarray analysis in relation to targetnumber
(A) Comparison of overall assay accuracy (both sensitivity and specificity) when screeningagainst an increasing number of target antigens Assay specificity declines as the numbersof target antigens increases Improved assay performance is achieved when a small panel of10 targets is selected (B) Dot plot compares the maximum binding intensity signal for eachpatient or control when comparing single and heteromeric complexes GBS patient medianarbitrary binding intensity for single targets (7473 FIU) is significantly less than for hetero-meric complex targets (35899 FIU p 00001) indicating that heteromeric complexesreturn higher signals than single antigens
6 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
GBS variants (n 5 105) were excluded from thisanalysis Six targets were identified as being signif-icant in the AIDP population when compared withthe AMAN clinical variant These included SGPGalone as well as 2 complexes containing SGPG(SGPGGQ1b and SGPGChol) In addition 2heteromeric complexes containing LM1 (LM1Sul-phatide and LM1Chol) and CTHChol were sig-nificantly associated with AIDP (p 005) Whilesignificant each of these targets was present at low
frequencies within the patient samples and thusindividually they had low diagnostic sensitivity(ranging from 105 to 140) but with a highspecificity both for combined controls (950) andpatients with AMAN (ranging from 960 to980 depending upon the target) When com-bined as a diagnostic panel of 6 antigen targets forAIDP a sensitivity of 339 and a specificity of873 and 834 (for AMAN and controls respec-tively) was reached
Table Glycolipid reactivities associated with acute inflammatory demyelinating polyneuropathy (AIDP)andacute motor axonal neuropathy (AMAN) clinical variants
AIDP (n 5 59) AMAN (n 5 102)
Guillain-Barreacute syndrome variant biomarker SGPG GM1 single 1 All GM1 complexes
SGPGGQ1b GA1 single 1 All GA1 complexes (excl GA1GQ1b)
SGPGChol GD1bGD1a
LM1Sulph GD1bPS
LM1Chol GD1bChol
CTHChol GD1bSulph
Ocular deficits GQ1bGD1b NA
GQ1bSGPG
GQ1bCTH
GQ1bChol
GQ1bGalC
LM1CTH
Bulbar palsy NA GQ1bPS
Facial palsy NA GA1GD1a
Figure 4 Heat map of antibody binding patterns in axonal vs demyelinating Guillain-Barreacute syndrome (GBS)cases
(A) Heat map presentation of the 26 glycolipid targets significant for the acute motor axonal neuropathy variant of GBScompared with the demyelinating (acute inflammatory demyelinating polyneuropathy) clinical subtype (B) Refining thescreening targets to a panel of 3 heteromeric complexes (GM1GQ1b GM1phosphatidylserine and GA1GD1a) resultsin a sensitivity of 784 and a specificity of 877 (for combined controls)
Neurology Neuroimmunology amp Neuroinflammation 7
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Twenty-six targets were identified as being sig-nificant (p 005) in AMAN when compared withthe AIDP population (figure 4A) These includeGM1 alone plus GM1 in complex with all testedlipids (GM1GA1 GM1GD1a GM1GD1bGM1GQ1b GM1SGPG GM1LM1 GM1CTH GM1PS GM1Chol GM1SulphatideGM1GalC) GA1 alone plus GA1 in complex withall tested lipids with the exception of GA1GQ1b(GA1GD1a GA1GD1b GA1SGPG GA1LM1GA1CTH GA1PS GA1Chol GA1SulphatideGA1GalC) GD1bGD1a GD1bPS GD1bCholand GD1bSulphatide Of these GM1GD1a hadthe highest sensitivity (559) and a high specific-ity for both combined controls (950) and pa-tients with AIDP (848) When combined asa diagnostic panel of 26 antigen targets for AMANa sensitivity of 863 and a specificity of 644and 458 (for combined controls and AIDPrespectively) was reached As previously demon-strated the specificity was reduced with increasingantigen targets therefore we sought to refine thenumber of antigen targets in order to optimize sen-sitivity and specificity When selecting only 3 anti-gen targets (GM1GQ1b GM1PS and GA1GD1a) a sensitivity of 784 and specificity of877 and 610 (for combined control andAIDP respectively) was achieved (figure 4B)
DISCUSSION This study applies technical develop-ments in antiglycolipid antibody assay miniaturizationto address the practical complexities of large-scalebiomarker screening of clinical cohorts By adaptingthe microarray printer to accommodate glycolipidswe were able to print arrays in a high throughputmanner One maximum capacity assay is capable ofprinting 320 individual arrays on a total of 20slides Conducting an equivalent screen on thissample size and antigen repertoire in a conventionalELISA study would require 640 ELISA plates (96-well) Miniaturization of the glycolipid assay allowsus to use 100-fold smaller volumes of patient serumthus the total serum use by microarray is 2 mL toprobe against 78 target antigens with duplicatemeasurement compared with 200 mL of serum fora comparable study performed on ELISA In additionthe use of fluorescently conjugated antibodies allowsfor multiplex screening whereby IgG andimmunoglobulin M can be detected simultaneouslyon each assay platform a benefit not routinelyavailable in ELISA readers
In this study a single electrophysiologic assess-ment was performed on 192 of the 266 patientsin the GBS cohort Of these patients 531 of pa-tients were classified as having an axonal variant ofGBS 302 of patients with AIDP while the
clinical variant of the remaining 167 of patientswas unclassified These diagnostic categorizationfrequencies are in line with a previously reportedGBS cohort from Bangladesh7 As found here thisprevious report also identified a low incidence ofpatient serum containing anti-GD1a antibodies(14) which was surprising given the high fre-quency of axonal GBS in comparison with studiesfrom elsewhere68ndash10 However probing Bangladeshsera against heteromeric complexes provedadvantageous as it revealed populations of complex-dependent GA1GD1a and GM1GD1a antibodieswhich were distinct from antibodies binding GD1a asa single antigen Similarly GM1GQ1b complex-dependent antibody binding was identified as a dis-tinct population in 119 of AMAN samples whencompared to GM1 and GQ1b alone Using a refinedpanel of only 3 antiglycolipid targets a high sensitivityand specificity was achieved for patients with AMANin this cohort
In addition to screening GBS patient samples 579combined control sera were examined This screenidentified variations in the baseline levels of antigan-glioside antibodies reactivities for each of the targetsWhen performing conventional ELISA the currentgold standard method for antiganglioside antibodydetection a universal optical density threshold of pos-itivity across all glycolipids is widely used In thisgeographically restricted study we can see that inorder to optimize both sensitivity and specificity ofthe assay individual threshold must be calculatedin order to determine the normal range of these nat-urally occurring and GBS-independent anticarbohy-drate antibodies
In this study we present data that identify thepresence of both single and heteromeric glycolipidcomplex binding antibodies in this large GBS cohortInitial biomarker screening employed a panel of 78targets however the selective inclusion of a very lim-ited panel of targets enabled optimization of both sen-sitivity and specificity due to significant targetredundancy resulting from the presence of polyclonalantibodies or cross-reactivity of specific antibodyspecies When considering the overwhelming com-plexities of identifying diagnostically important glyco-lipid complexes by ELISA screening the necessitiesof miniaturization are clearly evident for the initialanalysis in order to find the markers that are mostinformative
AUTHOR CONTRIBUTIONSSK Halstead study concept and design experimental work principal
draft of manuscript G Kalna statistical analysis and interpretation
MB Islam acquisition of clinical data I Jahan acquisition of clinical
data QD Mohammad acquisition of clinical data BC Jacobs study
concept and design study supervision HP Endtz study concept and
design study supervision Z Islam study concept and design critical
8 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
revision of the manuscript for important intellectual content HJ Willison
study concept and design critical revision of the manuscript for important
intellectual content study supervision
STUDY FUNDINGThis study is funded by the Wellcome Trust and the Chandra Mehta
Foundation
DISCLOSURESK Halstead G Kalna MB Islam I Jahan and QD Mohammad
report no disclosures BC Jacobs received travel funding from Baxter
International is on the editorial board for Journal of the Peripheral Ner-
vous System and received research support from Grifols CSL-Behring
Netherlands Organization for Health Research and Development Eras-
mus MC Prinses Beatrix GBS-CIDP Foundation International Prinses
Beatrix Fonds and Prinses Beatrix Spierfonds HP Endtz received
research support from EC Horizon 2020 Foundation Merieux Bill amp
Melinda Gates Foundation and GBS-CIDP foundation Z Islam reports
no disclosures HJ Willison is on the editorial board for Nature Clinical
Practice Neurology Journal of Neuroimmunology Muscle and Nerve Jour-
nal of Peripheral Nervous System Clinical and Experimental Neuroimmu-
nology and Experimental Neurology is an associate editor for BMC
Neurology holds a patent for combinatorial glycoarray technology and
received research support from Ipsen Annexon Alexion and MRC Go
to Neurologyorgnn for full disclosure forms
Received April 4 2016 Accepted in final form August 5 2016
REFERENCES1 Rinaldi S Brennan KM Willison HJ Heteromeric glyco-
lipid complexes as modulators of autoantibody and lectin
binding Prog Lipid Res 20104987ndash95
2 Kaida K Morita D Kanzaki M et al Ganglioside com-
plexes as new target antigens in Guillain-Barre syndrome
Ann Neurol 200456567ndash571
3 Galban-Horcajo F Halstead SK McGonigal R
Willison HJ The application of glycosphingolipid arrays
to autoantibody detection in neuroimmunological disor-
ders Curr Opin Chem Biol 20141878ndash86
4 Asbury AK Cornblath DR Assessment of current diag-
nostic criteria for Guillain-Barre syndrome Ann Neurol
1990(27 suppl)S21ndashS24
5 Jones CM Athanasiou T Summary receiver operating
characteristic curve analysis techniques in the evaluation
of diagnostic tests Ann Thorac Surg 20057916ndash20
6 Ho TW Willison HJ Nachamkin I et al Anti-GD1a
antibody is associated with axonal but not demyelinating
forms of Guillain-Barre syndrome Ann Neurol 199945
168ndash173
7 Islam Z Jacobs BC van Belkum A et al Axonal var-
iant of Guillain-Barre syndrome associated with cam-
pylobacter infection in Bangladesh Neurology 2010
74581ndash587
8 Kim JK Bae JS Kim DS et al Prevalence of anti-
ganglioside antibodies and their clinical correlates with
Guillain-Barreacute syndrome in Korea a nationwide multicen-
ter study J Clin Neurol 20141094ndash100
9 Willison HJ Yuki N Peripheral neuropathies and anti-
glycolipid antibodies Brain 20021252591ndash2625
10 Yuki N Yamada M Sato S et al Association of IgG anti-
GD1a antibody with severe Guillain-Barre syndrome
Muscle Nerve 199316642ndash647
Neurology Neuroimmunology amp Neuroinflammation 9
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
DOI 101212NXI000000000000028420163 Neurol Neuroimmunol Neuroinflamm
Susan K Halstead Gabriela Kalna Mohammad B Islam et al complexes
Microarray screening of Guillain-Barreacute syndrome sera for antibodies to glycolipid
This information is current as of September 28 2016
ServicesUpdated Information amp
httpnnneurologyorgcontent36e284fullhtmlincluding high resolution figures can be found at
Supplementary Material
httpnnneurologyorgcontentsuppl2017022136e284DC2 httpnnneurologyorgcontentsuppl2016092836e284DC1
Supplementary material can be found at
References httpnnneurologyorgcontent36e284fullhtmlref-list-1
This article cites 9 articles 0 of which you can access for free at
Subspecialty Collections
httpnnneurologyorgcgicollectionperipheral_neuropathyPeripheral neuropathy
httpnnneurologyorgcgicollectionguillainbarre_syndromeGuillain-Barre syndrome
httpnnneurologyorgcgicollectionautoimmune_diseasesAutoimmune diseasesfollowing 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
2016 American Academy of Neurology All rights reserved Online ISSN 2332-7812Published since April 2014 it is an open-access online-only continuous publication journal Copyright copy
is an official journal of the American Academy of NeurologyNeurol Neuroimmunol Neuroinflamm
complex most usually in a completely complex-dependent fashion (ie no binding is present toeither partner alone figure 2D green lines)Similarly when considering GM1 and GQ1bwhile some anti-GM1 and anti-GQ1b sera haveattenuated binding intensities when in GM1GQ1b complex there is a significantly largegroup of sera in which major enhancement
occurs in the presence of the complex (figure 2Ggreen lines)
The added value and limitations of screening for
heteromeric complexes To assess the overall impactof including glycolipid heteromeric complexes inthe screening platform we compared the number ofGBS samples classified as positive (ie returning anintensity value 95th percentile of the combinedcontrols) for single glycolipids and heteromeric com-plexes for the 10 frequently observed reactivities (fig-ure 2A) A total of 131 of 266 GBS samples and 87 of579 combined controls were positive for one or moresingle targets (sensitivity 492 specificity 850)whereas 155266 GBS samples and 95579 com-bined controls were positive for one or more hetero-meric complexes (sensitivity 583 specificity836) Therefore by screening for heteromericcomplexes a gain of 91 in sensitivity (p 5
00021) is achieved without a significant loss in spec-ificity (14 p 5 0445) Screening of GBS serairrespective of the clinical variant (AMAN AIDPor unclassified) against this discrete panel of 10 gly-colipids (4 single and 6 complexes) resulted in anoverall assay sensitivity of 643 and specificity of771
However when examining a larger panel of tar-gets a gain in sensitivity is frequently offset by anequivalent or greater loss in specificity (figure 3A)For example when examining the 30 targets withsignificantly different fluorescent intensity values (fig-ure 1B) a gain of 304 in sensitivity (from 410to 714) is offset by a loss of 29 specificity (from912 to 622) when comparing single and com-plex positivity frequency Similarly when consideringall 78 targets printed on the glycolipid arrays there isa 147 gain in sensitivity (from 650 to 797)while specificity drops 243 (from 658 to415) for single and complexes respectively Thesedata show that when examining the data in its total-ity the disease specificity of a very broad unbiasedscreen is expectedly poor owing to the large numberof controls whose sera contain at least one species ofantiglycolipid antibody above the normal range(95th percentile)
Antiglycolipid antibodies associated with clinical
variants Correlation between antiglycolipid anti-body profiles and clinical features are presented intable and as e-Results Having determined a panelof 10 glycolipid reactivities optimized for sensitivityand specificity for all patients with GBS irrespectiveof the clinical variant next we sought to determinewhether the presence of specific antibody reactiv-ities would segregate with a clinical variant or couldbe used to predict the clinical phenotype or out-come (see e-Results) Patients who had unclassified
Figure 3 Sensitivity and specificity of microarray analysis in relation to targetnumber
(A) Comparison of overall assay accuracy (both sensitivity and specificity) when screeningagainst an increasing number of target antigens Assay specificity declines as the numbersof target antigens increases Improved assay performance is achieved when a small panel of10 targets is selected (B) Dot plot compares the maximum binding intensity signal for eachpatient or control when comparing single and heteromeric complexes GBS patient medianarbitrary binding intensity for single targets (7473 FIU) is significantly less than for hetero-meric complex targets (35899 FIU p 00001) indicating that heteromeric complexesreturn higher signals than single antigens
6 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
GBS variants (n 5 105) were excluded from thisanalysis Six targets were identified as being signif-icant in the AIDP population when compared withthe AMAN clinical variant These included SGPGalone as well as 2 complexes containing SGPG(SGPGGQ1b and SGPGChol) In addition 2heteromeric complexes containing LM1 (LM1Sul-phatide and LM1Chol) and CTHChol were sig-nificantly associated with AIDP (p 005) Whilesignificant each of these targets was present at low
frequencies within the patient samples and thusindividually they had low diagnostic sensitivity(ranging from 105 to 140) but with a highspecificity both for combined controls (950) andpatients with AMAN (ranging from 960 to980 depending upon the target) When com-bined as a diagnostic panel of 6 antigen targets forAIDP a sensitivity of 339 and a specificity of873 and 834 (for AMAN and controls respec-tively) was reached
Table Glycolipid reactivities associated with acute inflammatory demyelinating polyneuropathy (AIDP)andacute motor axonal neuropathy (AMAN) clinical variants
AIDP (n 5 59) AMAN (n 5 102)
Guillain-Barreacute syndrome variant biomarker SGPG GM1 single 1 All GM1 complexes
SGPGGQ1b GA1 single 1 All GA1 complexes (excl GA1GQ1b)
SGPGChol GD1bGD1a
LM1Sulph GD1bPS
LM1Chol GD1bChol
CTHChol GD1bSulph
Ocular deficits GQ1bGD1b NA
GQ1bSGPG
GQ1bCTH
GQ1bChol
GQ1bGalC
LM1CTH
Bulbar palsy NA GQ1bPS
Facial palsy NA GA1GD1a
Figure 4 Heat map of antibody binding patterns in axonal vs demyelinating Guillain-Barreacute syndrome (GBS)cases
(A) Heat map presentation of the 26 glycolipid targets significant for the acute motor axonal neuropathy variant of GBScompared with the demyelinating (acute inflammatory demyelinating polyneuropathy) clinical subtype (B) Refining thescreening targets to a panel of 3 heteromeric complexes (GM1GQ1b GM1phosphatidylserine and GA1GD1a) resultsin a sensitivity of 784 and a specificity of 877 (for combined controls)
Neurology Neuroimmunology amp Neuroinflammation 7
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Twenty-six targets were identified as being sig-nificant (p 005) in AMAN when compared withthe AIDP population (figure 4A) These includeGM1 alone plus GM1 in complex with all testedlipids (GM1GA1 GM1GD1a GM1GD1bGM1GQ1b GM1SGPG GM1LM1 GM1CTH GM1PS GM1Chol GM1SulphatideGM1GalC) GA1 alone plus GA1 in complex withall tested lipids with the exception of GA1GQ1b(GA1GD1a GA1GD1b GA1SGPG GA1LM1GA1CTH GA1PS GA1Chol GA1SulphatideGA1GalC) GD1bGD1a GD1bPS GD1bCholand GD1bSulphatide Of these GM1GD1a hadthe highest sensitivity (559) and a high specific-ity for both combined controls (950) and pa-tients with AIDP (848) When combined asa diagnostic panel of 26 antigen targets for AMANa sensitivity of 863 and a specificity of 644and 458 (for combined controls and AIDPrespectively) was reached As previously demon-strated the specificity was reduced with increasingantigen targets therefore we sought to refine thenumber of antigen targets in order to optimize sen-sitivity and specificity When selecting only 3 anti-gen targets (GM1GQ1b GM1PS and GA1GD1a) a sensitivity of 784 and specificity of877 and 610 (for combined control andAIDP respectively) was achieved (figure 4B)
DISCUSSION This study applies technical develop-ments in antiglycolipid antibody assay miniaturizationto address the practical complexities of large-scalebiomarker screening of clinical cohorts By adaptingthe microarray printer to accommodate glycolipidswe were able to print arrays in a high throughputmanner One maximum capacity assay is capable ofprinting 320 individual arrays on a total of 20slides Conducting an equivalent screen on thissample size and antigen repertoire in a conventionalELISA study would require 640 ELISA plates (96-well) Miniaturization of the glycolipid assay allowsus to use 100-fold smaller volumes of patient serumthus the total serum use by microarray is 2 mL toprobe against 78 target antigens with duplicatemeasurement compared with 200 mL of serum fora comparable study performed on ELISA In additionthe use of fluorescently conjugated antibodies allowsfor multiplex screening whereby IgG andimmunoglobulin M can be detected simultaneouslyon each assay platform a benefit not routinelyavailable in ELISA readers
In this study a single electrophysiologic assess-ment was performed on 192 of the 266 patientsin the GBS cohort Of these patients 531 of pa-tients were classified as having an axonal variant ofGBS 302 of patients with AIDP while the
clinical variant of the remaining 167 of patientswas unclassified These diagnostic categorizationfrequencies are in line with a previously reportedGBS cohort from Bangladesh7 As found here thisprevious report also identified a low incidence ofpatient serum containing anti-GD1a antibodies(14) which was surprising given the high fre-quency of axonal GBS in comparison with studiesfrom elsewhere68ndash10 However probing Bangladeshsera against heteromeric complexes provedadvantageous as it revealed populations of complex-dependent GA1GD1a and GM1GD1a antibodieswhich were distinct from antibodies binding GD1a asa single antigen Similarly GM1GQ1b complex-dependent antibody binding was identified as a dis-tinct population in 119 of AMAN samples whencompared to GM1 and GQ1b alone Using a refinedpanel of only 3 antiglycolipid targets a high sensitivityand specificity was achieved for patients with AMANin this cohort
In addition to screening GBS patient samples 579combined control sera were examined This screenidentified variations in the baseline levels of antigan-glioside antibodies reactivities for each of the targetsWhen performing conventional ELISA the currentgold standard method for antiganglioside antibodydetection a universal optical density threshold of pos-itivity across all glycolipids is widely used In thisgeographically restricted study we can see that inorder to optimize both sensitivity and specificity ofthe assay individual threshold must be calculatedin order to determine the normal range of these nat-urally occurring and GBS-independent anticarbohy-drate antibodies
In this study we present data that identify thepresence of both single and heteromeric glycolipidcomplex binding antibodies in this large GBS cohortInitial biomarker screening employed a panel of 78targets however the selective inclusion of a very lim-ited panel of targets enabled optimization of both sen-sitivity and specificity due to significant targetredundancy resulting from the presence of polyclonalantibodies or cross-reactivity of specific antibodyspecies When considering the overwhelming com-plexities of identifying diagnostically important glyco-lipid complexes by ELISA screening the necessitiesof miniaturization are clearly evident for the initialanalysis in order to find the markers that are mostinformative
AUTHOR CONTRIBUTIONSSK Halstead study concept and design experimental work principal
draft of manuscript G Kalna statistical analysis and interpretation
MB Islam acquisition of clinical data I Jahan acquisition of clinical
data QD Mohammad acquisition of clinical data BC Jacobs study
concept and design study supervision HP Endtz study concept and
design study supervision Z Islam study concept and design critical
8 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
revision of the manuscript for important intellectual content HJ Willison
study concept and design critical revision of the manuscript for important
intellectual content study supervision
STUDY FUNDINGThis study is funded by the Wellcome Trust and the Chandra Mehta
Foundation
DISCLOSURESK Halstead G Kalna MB Islam I Jahan and QD Mohammad
report no disclosures BC Jacobs received travel funding from Baxter
International is on the editorial board for Journal of the Peripheral Ner-
vous System and received research support from Grifols CSL-Behring
Netherlands Organization for Health Research and Development Eras-
mus MC Prinses Beatrix GBS-CIDP Foundation International Prinses
Beatrix Fonds and Prinses Beatrix Spierfonds HP Endtz received
research support from EC Horizon 2020 Foundation Merieux Bill amp
Melinda Gates Foundation and GBS-CIDP foundation Z Islam reports
no disclosures HJ Willison is on the editorial board for Nature Clinical
Practice Neurology Journal of Neuroimmunology Muscle and Nerve Jour-
nal of Peripheral Nervous System Clinical and Experimental Neuroimmu-
nology and Experimental Neurology is an associate editor for BMC
Neurology holds a patent for combinatorial glycoarray technology and
received research support from Ipsen Annexon Alexion and MRC Go
to Neurologyorgnn for full disclosure forms
Received April 4 2016 Accepted in final form August 5 2016
REFERENCES1 Rinaldi S Brennan KM Willison HJ Heteromeric glyco-
lipid complexes as modulators of autoantibody and lectin
binding Prog Lipid Res 20104987ndash95
2 Kaida K Morita D Kanzaki M et al Ganglioside com-
plexes as new target antigens in Guillain-Barre syndrome
Ann Neurol 200456567ndash571
3 Galban-Horcajo F Halstead SK McGonigal R
Willison HJ The application of glycosphingolipid arrays
to autoantibody detection in neuroimmunological disor-
ders Curr Opin Chem Biol 20141878ndash86
4 Asbury AK Cornblath DR Assessment of current diag-
nostic criteria for Guillain-Barre syndrome Ann Neurol
1990(27 suppl)S21ndashS24
5 Jones CM Athanasiou T Summary receiver operating
characteristic curve analysis techniques in the evaluation
of diagnostic tests Ann Thorac Surg 20057916ndash20
6 Ho TW Willison HJ Nachamkin I et al Anti-GD1a
antibody is associated with axonal but not demyelinating
forms of Guillain-Barre syndrome Ann Neurol 199945
168ndash173
7 Islam Z Jacobs BC van Belkum A et al Axonal var-
iant of Guillain-Barre syndrome associated with cam-
pylobacter infection in Bangladesh Neurology 2010
74581ndash587
8 Kim JK Bae JS Kim DS et al Prevalence of anti-
ganglioside antibodies and their clinical correlates with
Guillain-Barreacute syndrome in Korea a nationwide multicen-
ter study J Clin Neurol 20141094ndash100
9 Willison HJ Yuki N Peripheral neuropathies and anti-
glycolipid antibodies Brain 20021252591ndash2625
10 Yuki N Yamada M Sato S et al Association of IgG anti-
GD1a antibody with severe Guillain-Barre syndrome
Muscle Nerve 199316642ndash647
Neurology Neuroimmunology amp Neuroinflammation 9
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
DOI 101212NXI000000000000028420163 Neurol Neuroimmunol Neuroinflamm
Susan K Halstead Gabriela Kalna Mohammad B Islam et al complexes
Microarray screening of Guillain-Barreacute syndrome sera for antibodies to glycolipid
This information is current as of September 28 2016
ServicesUpdated Information amp
httpnnneurologyorgcontent36e284fullhtmlincluding high resolution figures can be found at
Supplementary Material
httpnnneurologyorgcontentsuppl2017022136e284DC2 httpnnneurologyorgcontentsuppl2016092836e284DC1
Supplementary material can be found at
References httpnnneurologyorgcontent36e284fullhtmlref-list-1
This article cites 9 articles 0 of which you can access for free at
Subspecialty Collections
httpnnneurologyorgcgicollectionperipheral_neuropathyPeripheral neuropathy
httpnnneurologyorgcgicollectionguillainbarre_syndromeGuillain-Barre syndrome
httpnnneurologyorgcgicollectionautoimmune_diseasesAutoimmune diseasesfollowing 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
2016 American Academy of Neurology All rights reserved Online ISSN 2332-7812Published since April 2014 it is an open-access online-only continuous publication journal Copyright copy
is an official journal of the American Academy of NeurologyNeurol Neuroimmunol Neuroinflamm
GBS variants (n 5 105) were excluded from thisanalysis Six targets were identified as being signif-icant in the AIDP population when compared withthe AMAN clinical variant These included SGPGalone as well as 2 complexes containing SGPG(SGPGGQ1b and SGPGChol) In addition 2heteromeric complexes containing LM1 (LM1Sul-phatide and LM1Chol) and CTHChol were sig-nificantly associated with AIDP (p 005) Whilesignificant each of these targets was present at low
frequencies within the patient samples and thusindividually they had low diagnostic sensitivity(ranging from 105 to 140) but with a highspecificity both for combined controls (950) andpatients with AMAN (ranging from 960 to980 depending upon the target) When com-bined as a diagnostic panel of 6 antigen targets forAIDP a sensitivity of 339 and a specificity of873 and 834 (for AMAN and controls respec-tively) was reached
Table Glycolipid reactivities associated with acute inflammatory demyelinating polyneuropathy (AIDP)andacute motor axonal neuropathy (AMAN) clinical variants
AIDP (n 5 59) AMAN (n 5 102)
Guillain-Barreacute syndrome variant biomarker SGPG GM1 single 1 All GM1 complexes
SGPGGQ1b GA1 single 1 All GA1 complexes (excl GA1GQ1b)
SGPGChol GD1bGD1a
LM1Sulph GD1bPS
LM1Chol GD1bChol
CTHChol GD1bSulph
Ocular deficits GQ1bGD1b NA
GQ1bSGPG
GQ1bCTH
GQ1bChol
GQ1bGalC
LM1CTH
Bulbar palsy NA GQ1bPS
Facial palsy NA GA1GD1a
Figure 4 Heat map of antibody binding patterns in axonal vs demyelinating Guillain-Barreacute syndrome (GBS)cases
(A) Heat map presentation of the 26 glycolipid targets significant for the acute motor axonal neuropathy variant of GBScompared with the demyelinating (acute inflammatory demyelinating polyneuropathy) clinical subtype (B) Refining thescreening targets to a panel of 3 heteromeric complexes (GM1GQ1b GM1phosphatidylserine and GA1GD1a) resultsin a sensitivity of 784 and a specificity of 877 (for combined controls)
Neurology Neuroimmunology amp Neuroinflammation 7
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
Twenty-six targets were identified as being sig-nificant (p 005) in AMAN when compared withthe AIDP population (figure 4A) These includeGM1 alone plus GM1 in complex with all testedlipids (GM1GA1 GM1GD1a GM1GD1bGM1GQ1b GM1SGPG GM1LM1 GM1CTH GM1PS GM1Chol GM1SulphatideGM1GalC) GA1 alone plus GA1 in complex withall tested lipids with the exception of GA1GQ1b(GA1GD1a GA1GD1b GA1SGPG GA1LM1GA1CTH GA1PS GA1Chol GA1SulphatideGA1GalC) GD1bGD1a GD1bPS GD1bCholand GD1bSulphatide Of these GM1GD1a hadthe highest sensitivity (559) and a high specific-ity for both combined controls (950) and pa-tients with AIDP (848) When combined asa diagnostic panel of 26 antigen targets for AMANa sensitivity of 863 and a specificity of 644and 458 (for combined controls and AIDPrespectively) was reached As previously demon-strated the specificity was reduced with increasingantigen targets therefore we sought to refine thenumber of antigen targets in order to optimize sen-sitivity and specificity When selecting only 3 anti-gen targets (GM1GQ1b GM1PS and GA1GD1a) a sensitivity of 784 and specificity of877 and 610 (for combined control andAIDP respectively) was achieved (figure 4B)
DISCUSSION This study applies technical develop-ments in antiglycolipid antibody assay miniaturizationto address the practical complexities of large-scalebiomarker screening of clinical cohorts By adaptingthe microarray printer to accommodate glycolipidswe were able to print arrays in a high throughputmanner One maximum capacity assay is capable ofprinting 320 individual arrays on a total of 20slides Conducting an equivalent screen on thissample size and antigen repertoire in a conventionalELISA study would require 640 ELISA plates (96-well) Miniaturization of the glycolipid assay allowsus to use 100-fold smaller volumes of patient serumthus the total serum use by microarray is 2 mL toprobe against 78 target antigens with duplicatemeasurement compared with 200 mL of serum fora comparable study performed on ELISA In additionthe use of fluorescently conjugated antibodies allowsfor multiplex screening whereby IgG andimmunoglobulin M can be detected simultaneouslyon each assay platform a benefit not routinelyavailable in ELISA readers
In this study a single electrophysiologic assess-ment was performed on 192 of the 266 patientsin the GBS cohort Of these patients 531 of pa-tients were classified as having an axonal variant ofGBS 302 of patients with AIDP while the
clinical variant of the remaining 167 of patientswas unclassified These diagnostic categorizationfrequencies are in line with a previously reportedGBS cohort from Bangladesh7 As found here thisprevious report also identified a low incidence ofpatient serum containing anti-GD1a antibodies(14) which was surprising given the high fre-quency of axonal GBS in comparison with studiesfrom elsewhere68ndash10 However probing Bangladeshsera against heteromeric complexes provedadvantageous as it revealed populations of complex-dependent GA1GD1a and GM1GD1a antibodieswhich were distinct from antibodies binding GD1a asa single antigen Similarly GM1GQ1b complex-dependent antibody binding was identified as a dis-tinct population in 119 of AMAN samples whencompared to GM1 and GQ1b alone Using a refinedpanel of only 3 antiglycolipid targets a high sensitivityand specificity was achieved for patients with AMANin this cohort
In addition to screening GBS patient samples 579combined control sera were examined This screenidentified variations in the baseline levels of antigan-glioside antibodies reactivities for each of the targetsWhen performing conventional ELISA the currentgold standard method for antiganglioside antibodydetection a universal optical density threshold of pos-itivity across all glycolipids is widely used In thisgeographically restricted study we can see that inorder to optimize both sensitivity and specificity ofthe assay individual threshold must be calculatedin order to determine the normal range of these nat-urally occurring and GBS-independent anticarbohy-drate antibodies
In this study we present data that identify thepresence of both single and heteromeric glycolipidcomplex binding antibodies in this large GBS cohortInitial biomarker screening employed a panel of 78targets however the selective inclusion of a very lim-ited panel of targets enabled optimization of both sen-sitivity and specificity due to significant targetredundancy resulting from the presence of polyclonalantibodies or cross-reactivity of specific antibodyspecies When considering the overwhelming com-plexities of identifying diagnostically important glyco-lipid complexes by ELISA screening the necessitiesof miniaturization are clearly evident for the initialanalysis in order to find the markers that are mostinformative
AUTHOR CONTRIBUTIONSSK Halstead study concept and design experimental work principal
draft of manuscript G Kalna statistical analysis and interpretation
MB Islam acquisition of clinical data I Jahan acquisition of clinical
data QD Mohammad acquisition of clinical data BC Jacobs study
concept and design study supervision HP Endtz study concept and
design study supervision Z Islam study concept and design critical
8 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
revision of the manuscript for important intellectual content HJ Willison
study concept and design critical revision of the manuscript for important
intellectual content study supervision
STUDY FUNDINGThis study is funded by the Wellcome Trust and the Chandra Mehta
Foundation
DISCLOSURESK Halstead G Kalna MB Islam I Jahan and QD Mohammad
report no disclosures BC Jacobs received travel funding from Baxter
International is on the editorial board for Journal of the Peripheral Ner-
vous System and received research support from Grifols CSL-Behring
Netherlands Organization for Health Research and Development Eras-
mus MC Prinses Beatrix GBS-CIDP Foundation International Prinses
Beatrix Fonds and Prinses Beatrix Spierfonds HP Endtz received
research support from EC Horizon 2020 Foundation Merieux Bill amp
Melinda Gates Foundation and GBS-CIDP foundation Z Islam reports
no disclosures HJ Willison is on the editorial board for Nature Clinical
Practice Neurology Journal of Neuroimmunology Muscle and Nerve Jour-
nal of Peripheral Nervous System Clinical and Experimental Neuroimmu-
nology and Experimental Neurology is an associate editor for BMC
Neurology holds a patent for combinatorial glycoarray technology and
received research support from Ipsen Annexon Alexion and MRC Go
to Neurologyorgnn for full disclosure forms
Received April 4 2016 Accepted in final form August 5 2016
REFERENCES1 Rinaldi S Brennan KM Willison HJ Heteromeric glyco-
lipid complexes as modulators of autoantibody and lectin
binding Prog Lipid Res 20104987ndash95
2 Kaida K Morita D Kanzaki M et al Ganglioside com-
plexes as new target antigens in Guillain-Barre syndrome
Ann Neurol 200456567ndash571
3 Galban-Horcajo F Halstead SK McGonigal R
Willison HJ The application of glycosphingolipid arrays
to autoantibody detection in neuroimmunological disor-
ders Curr Opin Chem Biol 20141878ndash86
4 Asbury AK Cornblath DR Assessment of current diag-
nostic criteria for Guillain-Barre syndrome Ann Neurol
1990(27 suppl)S21ndashS24
5 Jones CM Athanasiou T Summary receiver operating
characteristic curve analysis techniques in the evaluation
of diagnostic tests Ann Thorac Surg 20057916ndash20
6 Ho TW Willison HJ Nachamkin I et al Anti-GD1a
antibody is associated with axonal but not demyelinating
forms of Guillain-Barre syndrome Ann Neurol 199945
168ndash173
7 Islam Z Jacobs BC van Belkum A et al Axonal var-
iant of Guillain-Barre syndrome associated with cam-
pylobacter infection in Bangladesh Neurology 2010
74581ndash587
8 Kim JK Bae JS Kim DS et al Prevalence of anti-
ganglioside antibodies and their clinical correlates with
Guillain-Barreacute syndrome in Korea a nationwide multicen-
ter study J Clin Neurol 20141094ndash100
9 Willison HJ Yuki N Peripheral neuropathies and anti-
glycolipid antibodies Brain 20021252591ndash2625
10 Yuki N Yamada M Sato S et al Association of IgG anti-
GD1a antibody with severe Guillain-Barre syndrome
Muscle Nerve 199316642ndash647
Neurology Neuroimmunology amp Neuroinflammation 9
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
DOI 101212NXI000000000000028420163 Neurol Neuroimmunol Neuroinflamm
Susan K Halstead Gabriela Kalna Mohammad B Islam et al complexes
Microarray screening of Guillain-Barreacute syndrome sera for antibodies to glycolipid
This information is current as of September 28 2016
ServicesUpdated Information amp
httpnnneurologyorgcontent36e284fullhtmlincluding high resolution figures can be found at
Supplementary Material
httpnnneurologyorgcontentsuppl2017022136e284DC2 httpnnneurologyorgcontentsuppl2016092836e284DC1
Supplementary material can be found at
References httpnnneurologyorgcontent36e284fullhtmlref-list-1
This article cites 9 articles 0 of which you can access for free at
Subspecialty Collections
httpnnneurologyorgcgicollectionperipheral_neuropathyPeripheral neuropathy
httpnnneurologyorgcgicollectionguillainbarre_syndromeGuillain-Barre syndrome
httpnnneurologyorgcgicollectionautoimmune_diseasesAutoimmune diseasesfollowing 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
2016 American Academy of Neurology All rights reserved Online ISSN 2332-7812Published since April 2014 it is an open-access online-only continuous publication journal Copyright copy
is an official journal of the American Academy of NeurologyNeurol Neuroimmunol Neuroinflamm
Twenty-six targets were identified as being sig-nificant (p 005) in AMAN when compared withthe AIDP population (figure 4A) These includeGM1 alone plus GM1 in complex with all testedlipids (GM1GA1 GM1GD1a GM1GD1bGM1GQ1b GM1SGPG GM1LM1 GM1CTH GM1PS GM1Chol GM1SulphatideGM1GalC) GA1 alone plus GA1 in complex withall tested lipids with the exception of GA1GQ1b(GA1GD1a GA1GD1b GA1SGPG GA1LM1GA1CTH GA1PS GA1Chol GA1SulphatideGA1GalC) GD1bGD1a GD1bPS GD1bCholand GD1bSulphatide Of these GM1GD1a hadthe highest sensitivity (559) and a high specific-ity for both combined controls (950) and pa-tients with AIDP (848) When combined asa diagnostic panel of 26 antigen targets for AMANa sensitivity of 863 and a specificity of 644and 458 (for combined controls and AIDPrespectively) was reached As previously demon-strated the specificity was reduced with increasingantigen targets therefore we sought to refine thenumber of antigen targets in order to optimize sen-sitivity and specificity When selecting only 3 anti-gen targets (GM1GQ1b GM1PS and GA1GD1a) a sensitivity of 784 and specificity of877 and 610 (for combined control andAIDP respectively) was achieved (figure 4B)
DISCUSSION This study applies technical develop-ments in antiglycolipid antibody assay miniaturizationto address the practical complexities of large-scalebiomarker screening of clinical cohorts By adaptingthe microarray printer to accommodate glycolipidswe were able to print arrays in a high throughputmanner One maximum capacity assay is capable ofprinting 320 individual arrays on a total of 20slides Conducting an equivalent screen on thissample size and antigen repertoire in a conventionalELISA study would require 640 ELISA plates (96-well) Miniaturization of the glycolipid assay allowsus to use 100-fold smaller volumes of patient serumthus the total serum use by microarray is 2 mL toprobe against 78 target antigens with duplicatemeasurement compared with 200 mL of serum fora comparable study performed on ELISA In additionthe use of fluorescently conjugated antibodies allowsfor multiplex screening whereby IgG andimmunoglobulin M can be detected simultaneouslyon each assay platform a benefit not routinelyavailable in ELISA readers
In this study a single electrophysiologic assess-ment was performed on 192 of the 266 patientsin the GBS cohort Of these patients 531 of pa-tients were classified as having an axonal variant ofGBS 302 of patients with AIDP while the
clinical variant of the remaining 167 of patientswas unclassified These diagnostic categorizationfrequencies are in line with a previously reportedGBS cohort from Bangladesh7 As found here thisprevious report also identified a low incidence ofpatient serum containing anti-GD1a antibodies(14) which was surprising given the high fre-quency of axonal GBS in comparison with studiesfrom elsewhere68ndash10 However probing Bangladeshsera against heteromeric complexes provedadvantageous as it revealed populations of complex-dependent GA1GD1a and GM1GD1a antibodieswhich were distinct from antibodies binding GD1a asa single antigen Similarly GM1GQ1b complex-dependent antibody binding was identified as a dis-tinct population in 119 of AMAN samples whencompared to GM1 and GQ1b alone Using a refinedpanel of only 3 antiglycolipid targets a high sensitivityand specificity was achieved for patients with AMANin this cohort
In addition to screening GBS patient samples 579combined control sera were examined This screenidentified variations in the baseline levels of antigan-glioside antibodies reactivities for each of the targetsWhen performing conventional ELISA the currentgold standard method for antiganglioside antibodydetection a universal optical density threshold of pos-itivity across all glycolipids is widely used In thisgeographically restricted study we can see that inorder to optimize both sensitivity and specificity ofthe assay individual threshold must be calculatedin order to determine the normal range of these nat-urally occurring and GBS-independent anticarbohy-drate antibodies
In this study we present data that identify thepresence of both single and heteromeric glycolipidcomplex binding antibodies in this large GBS cohortInitial biomarker screening employed a panel of 78targets however the selective inclusion of a very lim-ited panel of targets enabled optimization of both sen-sitivity and specificity due to significant targetredundancy resulting from the presence of polyclonalantibodies or cross-reactivity of specific antibodyspecies When considering the overwhelming com-plexities of identifying diagnostically important glyco-lipid complexes by ELISA screening the necessitiesof miniaturization are clearly evident for the initialanalysis in order to find the markers that are mostinformative
AUTHOR CONTRIBUTIONSSK Halstead study concept and design experimental work principal
draft of manuscript G Kalna statistical analysis and interpretation
MB Islam acquisition of clinical data I Jahan acquisition of clinical
data QD Mohammad acquisition of clinical data BC Jacobs study
concept and design study supervision HP Endtz study concept and
design study supervision Z Islam study concept and design critical
8 Neurology Neuroimmunology amp Neuroinflammation
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
revision of the manuscript for important intellectual content HJ Willison
study concept and design critical revision of the manuscript for important
intellectual content study supervision
STUDY FUNDINGThis study is funded by the Wellcome Trust and the Chandra Mehta
Foundation
DISCLOSURESK Halstead G Kalna MB Islam I Jahan and QD Mohammad
report no disclosures BC Jacobs received travel funding from Baxter
International is on the editorial board for Journal of the Peripheral Ner-
vous System and received research support from Grifols CSL-Behring
Netherlands Organization for Health Research and Development Eras-
mus MC Prinses Beatrix GBS-CIDP Foundation International Prinses
Beatrix Fonds and Prinses Beatrix Spierfonds HP Endtz received
research support from EC Horizon 2020 Foundation Merieux Bill amp
Melinda Gates Foundation and GBS-CIDP foundation Z Islam reports
no disclosures HJ Willison is on the editorial board for Nature Clinical
Practice Neurology Journal of Neuroimmunology Muscle and Nerve Jour-
nal of Peripheral Nervous System Clinical and Experimental Neuroimmu-
nology and Experimental Neurology is an associate editor for BMC
Neurology holds a patent for combinatorial glycoarray technology and
received research support from Ipsen Annexon Alexion and MRC Go
to Neurologyorgnn for full disclosure forms
Received April 4 2016 Accepted in final form August 5 2016
REFERENCES1 Rinaldi S Brennan KM Willison HJ Heteromeric glyco-
lipid complexes as modulators of autoantibody and lectin
binding Prog Lipid Res 20104987ndash95
2 Kaida K Morita D Kanzaki M et al Ganglioside com-
plexes as new target antigens in Guillain-Barre syndrome
Ann Neurol 200456567ndash571
3 Galban-Horcajo F Halstead SK McGonigal R
Willison HJ The application of glycosphingolipid arrays
to autoantibody detection in neuroimmunological disor-
ders Curr Opin Chem Biol 20141878ndash86
4 Asbury AK Cornblath DR Assessment of current diag-
nostic criteria for Guillain-Barre syndrome Ann Neurol
1990(27 suppl)S21ndashS24
5 Jones CM Athanasiou T Summary receiver operating
characteristic curve analysis techniques in the evaluation
of diagnostic tests Ann Thorac Surg 20057916ndash20
6 Ho TW Willison HJ Nachamkin I et al Anti-GD1a
antibody is associated with axonal but not demyelinating
forms of Guillain-Barre syndrome Ann Neurol 199945
168ndash173
7 Islam Z Jacobs BC van Belkum A et al Axonal var-
iant of Guillain-Barre syndrome associated with cam-
pylobacter infection in Bangladesh Neurology 2010
74581ndash587
8 Kim JK Bae JS Kim DS et al Prevalence of anti-
ganglioside antibodies and their clinical correlates with
Guillain-Barreacute syndrome in Korea a nationwide multicen-
ter study J Clin Neurol 20141094ndash100
9 Willison HJ Yuki N Peripheral neuropathies and anti-
glycolipid antibodies Brain 20021252591ndash2625
10 Yuki N Yamada M Sato S et al Association of IgG anti-
GD1a antibody with severe Guillain-Barre syndrome
Muscle Nerve 199316642ndash647
Neurology Neuroimmunology amp Neuroinflammation 9
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
DOI 101212NXI000000000000028420163 Neurol Neuroimmunol Neuroinflamm
Susan K Halstead Gabriela Kalna Mohammad B Islam et al complexes
Microarray screening of Guillain-Barreacute syndrome sera for antibodies to glycolipid
This information is current as of September 28 2016
ServicesUpdated Information amp
httpnnneurologyorgcontent36e284fullhtmlincluding high resolution figures can be found at
Supplementary Material
httpnnneurologyorgcontentsuppl2017022136e284DC2 httpnnneurologyorgcontentsuppl2016092836e284DC1
Supplementary material can be found at
References httpnnneurologyorgcontent36e284fullhtmlref-list-1
This article cites 9 articles 0 of which you can access for free at
Subspecialty Collections
httpnnneurologyorgcgicollectionperipheral_neuropathyPeripheral neuropathy
httpnnneurologyorgcgicollectionguillainbarre_syndromeGuillain-Barre syndrome
httpnnneurologyorgcgicollectionautoimmune_diseasesAutoimmune diseasesfollowing 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
2016 American Academy of Neurology All rights reserved Online ISSN 2332-7812Published since April 2014 it is an open-access online-only continuous publication journal Copyright copy
is an official journal of the American Academy of NeurologyNeurol Neuroimmunol Neuroinflamm
revision of the manuscript for important intellectual content HJ Willison
study concept and design critical revision of the manuscript for important
intellectual content study supervision
STUDY FUNDINGThis study is funded by the Wellcome Trust and the Chandra Mehta
Foundation
DISCLOSURESK Halstead G Kalna MB Islam I Jahan and QD Mohammad
report no disclosures BC Jacobs received travel funding from Baxter
International is on the editorial board for Journal of the Peripheral Ner-
vous System and received research support from Grifols CSL-Behring
Netherlands Organization for Health Research and Development Eras-
mus MC Prinses Beatrix GBS-CIDP Foundation International Prinses
Beatrix Fonds and Prinses Beatrix Spierfonds HP Endtz received
research support from EC Horizon 2020 Foundation Merieux Bill amp
Melinda Gates Foundation and GBS-CIDP foundation Z Islam reports
no disclosures HJ Willison is on the editorial board for Nature Clinical
Practice Neurology Journal of Neuroimmunology Muscle and Nerve Jour-
nal of Peripheral Nervous System Clinical and Experimental Neuroimmu-
nology and Experimental Neurology is an associate editor for BMC
Neurology holds a patent for combinatorial glycoarray technology and
received research support from Ipsen Annexon Alexion and MRC Go
to Neurologyorgnn for full disclosure forms
Received April 4 2016 Accepted in final form August 5 2016
REFERENCES1 Rinaldi S Brennan KM Willison HJ Heteromeric glyco-
lipid complexes as modulators of autoantibody and lectin
binding Prog Lipid Res 20104987ndash95
2 Kaida K Morita D Kanzaki M et al Ganglioside com-
plexes as new target antigens in Guillain-Barre syndrome
Ann Neurol 200456567ndash571
3 Galban-Horcajo F Halstead SK McGonigal R
Willison HJ The application of glycosphingolipid arrays
to autoantibody detection in neuroimmunological disor-
ders Curr Opin Chem Biol 20141878ndash86
4 Asbury AK Cornblath DR Assessment of current diag-
nostic criteria for Guillain-Barre syndrome Ann Neurol
1990(27 suppl)S21ndashS24
5 Jones CM Athanasiou T Summary receiver operating
characteristic curve analysis techniques in the evaluation
of diagnostic tests Ann Thorac Surg 20057916ndash20
6 Ho TW Willison HJ Nachamkin I et al Anti-GD1a
antibody is associated with axonal but not demyelinating
forms of Guillain-Barre syndrome Ann Neurol 199945
168ndash173
7 Islam Z Jacobs BC van Belkum A et al Axonal var-
iant of Guillain-Barre syndrome associated with cam-
pylobacter infection in Bangladesh Neurology 2010
74581ndash587
8 Kim JK Bae JS Kim DS et al Prevalence of anti-
ganglioside antibodies and their clinical correlates with
Guillain-Barreacute syndrome in Korea a nationwide multicen-
ter study J Clin Neurol 20141094ndash100
9 Willison HJ Yuki N Peripheral neuropathies and anti-
glycolipid antibodies Brain 20021252591ndash2625
10 Yuki N Yamada M Sato S et al Association of IgG anti-
GD1a antibody with severe Guillain-Barre syndrome
Muscle Nerve 199316642ndash647
Neurology Neuroimmunology amp Neuroinflammation 9
ordf 2016 American Academy of Neurology Unauthorized reproduction of this article is prohibited
DOI 101212NXI000000000000028420163 Neurol Neuroimmunol Neuroinflamm
Susan K Halstead Gabriela Kalna Mohammad B Islam et al complexes
Microarray screening of Guillain-Barreacute syndrome sera for antibodies to glycolipid
This information is current as of September 28 2016
ServicesUpdated Information amp
httpnnneurologyorgcontent36e284fullhtmlincluding high resolution figures can be found at
Supplementary Material
httpnnneurologyorgcontentsuppl2017022136e284DC2 httpnnneurologyorgcontentsuppl2016092836e284DC1
Supplementary material can be found at
References httpnnneurologyorgcontent36e284fullhtmlref-list-1
This article cites 9 articles 0 of which you can access for free at
Subspecialty Collections
httpnnneurologyorgcgicollectionperipheral_neuropathyPeripheral neuropathy
httpnnneurologyorgcgicollectionguillainbarre_syndromeGuillain-Barre syndrome
httpnnneurologyorgcgicollectionautoimmune_diseasesAutoimmune diseasesfollowing 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
2016 American Academy of Neurology All rights reserved Online ISSN 2332-7812Published since April 2014 it is an open-access online-only continuous publication journal Copyright copy
is an official journal of the American Academy of NeurologyNeurol Neuroimmunol Neuroinflamm
DOI 101212NXI000000000000028420163 Neurol Neuroimmunol Neuroinflamm
Susan K Halstead Gabriela Kalna Mohammad B Islam et al complexes
Microarray screening of Guillain-Barreacute syndrome sera for antibodies to glycolipid
This information is current as of September 28 2016
ServicesUpdated Information amp
httpnnneurologyorgcontent36e284fullhtmlincluding high resolution figures can be found at
Supplementary Material
httpnnneurologyorgcontentsuppl2017022136e284DC2 httpnnneurologyorgcontentsuppl2016092836e284DC1
Supplementary material can be found at
References httpnnneurologyorgcontent36e284fullhtmlref-list-1
This article cites 9 articles 0 of which you can access for free at
Subspecialty Collections
httpnnneurologyorgcgicollectionperipheral_neuropathyPeripheral neuropathy
httpnnneurologyorgcgicollectionguillainbarre_syndromeGuillain-Barre syndrome
httpnnneurologyorgcgicollectionautoimmune_diseasesAutoimmune diseasesfollowing 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
2016 American Academy of Neurology All rights reserved Online ISSN 2332-7812Published since April 2014 it is an open-access online-only continuous publication journal Copyright copy
is an official journal of the American Academy of NeurologyNeurol Neuroimmunol Neuroinflamm