Identification of novel candidate protein biomarkers for ...bme.lth.se/fileadmin/user_upload/Publications/09_JProt_Gonzalez.pdf · Identification of novel candidate protein biomarkers

J O U R N A L O F P R O T E O M I C S X X ( 2 0 0 9 ) X X X – X X X

ava i l ab l e a t www.sc i enced i rec t . com

www.e l sev i e r. com/ loca te / j p ro t

ARTICLE IN PRESSJPROT-00078; No of Pages 12

Identification of novel candidate protein biomarkers for thepost-polio syndrome — Implications for diagnosis,neurodegeneration and neuroinflammation

Henrik Gonzaleza,1, Jan Ottervaldb,f,⁎,1, Kerstin C. Nilssonc, Niclas Sjögrend,Tasso Miliotise, Helena Von Bahre, Mohsen Khademif, Bodil Erikssong, Sven Kjellströmh,Akos Vegvarih, Robert Harrisf, György Marko-Vargah, Kristian Borga, Johan Nilssoni,Thomas Laurelli, Tomas Olssonf,1, Bo Franzénb,1

aDivision of Rehabilitation Medicine, Department of Clinical Sciences, Danderyd Hospital, Karolinska Institute, Stockholm, SwedenbMolecular Pharmacology, AstraZeneca R&D Södertälje, SwedencDisease Biology, Local Discovery, AstraZeneca R&D Södertälje, SwedendBiostatistics, AstraZeneca R&D Södertälje, SwedeneBioscience, Local Discovery, AstraZeneca R&D Mölndal, SwedenfNeuroimmunology Unit, Department of Clinical Neuroscience, Karolinska Hospital, Stockholm, SwedengBioPR&D, AstraZeneca R&D Södertälje, SwedenhBiological Sciences, Local Discovery, AstraZeneca R&D Lund, SwedeniLund Technical University, Lund, Sweden

A R T I C L E D A T A

Abbreviations: PPS, Post-polio syndrome; COND, other neurological disease; PLS, Partiaamidating monooxygenase; Mr/pI, molecula⁎ Corresponding author. Neuroimmunology U

Sweden. Tel.: +46 8 51776246, +46 73 6459165E-mail address: [email protected] (J. Otte

1 Equal contribution.

1874-3919/$ – see front matter © 2008 Elsevidoi:10.1016/j.jprot.2008.11.014

Please cite this article as: Gonzalez H, etImplications for diagnosis, neurodegener

A B S T R A C T

Article history:Received 1 October 2008Accepted 14 November 2008

Survivors of poliomyelitis often develop increased or new symptoms decades after the acuteinfection, a condition known as post-polio syndrome (PPS). The condition affects 20–60% ofprevious polio patients, making it one of the most common causes of neurological deficitsworldwide. The underlying pathogenesis is not fully understood and accurate diagnosis isnot feasible. Herein we investigated whether it was possible to identify proteomic profileaberrations in the cerebrospinal fluid (CSF) of PPS patients.CSF from 15 patients with well-defined PPS were analyzed for protein expression profiles.The results were compared to data obtained from nine healthy controls and 34 patients withother non-inflammatory diseases which served as negative controls. In addition, 17 samplesfrom persons with secondary progressive multiple sclerosis (SPMS) were added as relevantage-matched references for the PPS samples.The CSF of persons with PPS displayed a disease-specific and highly predictive (p=0.0017)differential expression of five distinct proteins: gelsolin, hemopexin, peptidylglycine alpha-amidating monooxygenase, glutathione synthetase and kallikrein 6, respectively, incomparison with the control groups. An independent ELISA confirmed the increase ofkallikrein 6.

Keywords:Post-polio syndromePathophysiologyCSFProteomicsBiomarkersDiagnosis

SF, cerebrospinal fluid; SPMS, secondary progressive multiple sclerosis; HC, healthy controls;l Least Squares; VIP, Variable Importance in the Projection; PAM, peptidylglycine alpha-r weight/isoelectric point.nit, Centre for Molecular Medicine, CMM, L8:04, Karolinska Hospital, 171 76, Stockholm,; fax: +46 8 51776248.rvald).

er B.V. All rights reserved.

al, Identification of novel candidate protein biomarkers for the post-polio syndrome—ation and neuroinflammation, J Prot (2009), doi:10.1016/j.jprot.2008.11.014

mailto:[email protected]

http://dx.doi.org/10.1016/j.jprot.2008.11.014


2 J O U R N A L O F P R O T E O M I C S X X ( 2 0 0 9 ) X X X – X X X

ARTICLE IN PRESS

Please cite this article as: Gonzalez H, etImplications for diagnosis, neurodegener

We suggest that these five proteins should be further evaluated as candidate biomarkers forthe diagnosis and development of new therapies for PPS patients.

© 2008 Elsevier B.V. All rights reserved.

Table 1 – The diagnosis and number of patients used inthe study

Diagnosis Number Male Female Meanagemale

Meanage

female

Healthy controls⁎ 9 0 10 – 32.4OND⁎ 34 12 22 42.1 38.6Non-inflammatory

1. Introduction

The post-polio syndrome (PPS) appears later in life after anacute poliomyelitis infection with a prevalence of 20–60%.Approximately twentymillion persons worldwide are affectedto some degree. This makes PPS one of the most prevalentmotor neuron diseases [1]. Survivors of acute paralytic diseaselater in life develop symptoms such as increased weakness ofthe musculoskeletal system, fatigue, pain, gait disturbances,breathing difficulties as well as swallowing problems [2].Consequently, it may be difficult to differentiate symptoms ofan active PPS disease from those of normal aging. The currentdiagnosis is based on thorough clinical examinations in orderto eliminate other possible causes, and thus identification ofdiagnostic biomarkers is highly desirable. Furthermore, iden-tification of any protein abnormalities may give insight of theunderlying patophysiology, which in turn could indicate newtherapeutic options.

An ongoing denervation has been suggested as thecause of the increased weakness in muscles earlier affectedby poliomyelitis [3,4]. This denervation is compensated byreinnervation by means of collateral sprouting, leading toan increase of the area of the motor units. However,reinnervation cannot sufficiently compensate for denerva-tion, leading to whole or partial loss of motor unitsfollowed by a decrease of muscle strength [4]. It is debatedwhether this deterioration is due to further loss of motorneurons due to normal aging, deleterious over-use ofremaining motor neurons, and/or an active disease process,perhaps involving chronic intrathecal inflammatorydamage [5–19].

To study the patophysiology of PPS, CSF is of particularinterest since its protein content is expected to reflectchanges in the target organ. Previous studies of CSFproteins have suggested associated biomarkers in theneuroscience fields of Creutzfeldts disease [20] and Alzhei-mers disease [21,22]. In the present study we appliedclassical proteomics to take an unbiased approach tostudy a broad spectrum of proteins in the CSF from PPSpatients and control subjects with or without other centralnervous system disease. The aim was to identify candidatebiomarkers for PPS and further to uncover hithertounknown biological processes related to PPS.

SPMS⁎ 17 6 11 54.1 56.5Inflammatory

PPS⁎ 15 9 6 59.3 60.5OND ELISAevaluation

30 9 21 37.7 43

Non-inflammatoryPPS ELISAevaluation

37 16 21 61.5 56.9

Total number ofpatients

n=142 52 90 Totalmean

Totalmean

50.9 47.9

Patients included in the proteomic part indicated with ⁎.

2. Methods

2.1. Patient population

Consecutive PPS patients with a history of acute poliomye-litis and new problems such as increased muscle weakness,muscle fatigue and pain in muscle groups earlier affected bypoliomyelitis according to the criteria for PPS as given byHalstead and Rossi [23] were included. (Supplementary data

al, Identification of noveation and neuroinflamm

1.1) All 15 PPS patients in this study had been examinedbefore with regard to CSF mononuclear expression of TNF-αand IFN-γ and all displayed elevated mRNA expression inCSF of these cytokines as compared to in controls [24](Table 1). Control patient CSF samples were either from non-diseased volunteers, healthy controls (HC), or persons withneurological signs or symptoms for other reasons than PPS,‘other neurological disease’ (OND), respectively. In thefollowing text both healthy controls and OND patients aregrouped together as ‘OND/HC’ n=43. These patients wereregarded as non-inflammatory and the lumbar puncturehad been performed as a part of routine clinical analysis ofthese patients. All samples were assessed for erythrocytecontamination before analysis. The CSF of these patientsdisplayed no signs of inflammation in the form of pleocy-tosis, increased IgG index or oligoclonal bands. A compar-ison with samples from persons with secondary progressiveMultiple Sclerosis (SPMS) was added (n=17). SPMS sampleswere age-matched to the PPS group and can be considered tobe an inflammatory disease reference material (Table 1).

CSF from an independent population of persons with PPSand ONDwere used to validate Kallikrein 6 expression levels.CSF was obtained from 37 individuals with PPS and 30individuals diagnosed as being OND (Table 1).

2.2. Ethics

All study enrolment followed the recommendations of thedeclaration of Helsinki and the study was approved by theEthics committee of the Karolinska Institute #02-365 and #02-213. Oral and written information was given to the patientsand confirmed consent in writing was received beforeinclusion.

l candidate protein biomarkers for the post-polio syndrome—ation, J Prot (2009), doi:10.1016/j.jprot.2008.11.014


3J O U R N A L O F P R O T E O M I C S X X ( 2 0 0 9 ) X X X – X X X

ARTICLE IN PRESS

2.3. Sample preparation and protein separation

All samples were albumin- and IgG-depleted to increase thesensitivity and resolution of the analyses. Preparation andprotein concentration prior to 2-DE is described in Supple-mentary data 2. Samples was analyzed by 2-DE (isoelectricfocusing followed by SDS-PAGE) as described elsewhere [25],thereafter stained with fluorescent dye [26] and finallydigitized with a laser scanner. All samples were analyzed ina randomized sequence to avoid batch-to-batch influence ofthe results. Image analysis and quantitative protein detec-tion were made prior to analysis using PDQuest™ software(version 7.3, Bio-Rad, Hercules, CA, USA). The Mr/pI coordi-nates were calculated from previously known and identifiedproteins from CSF (http://expasy.org/swiss-2dpage/viewer.)

2.4. Data analysis

Image analysis of the 2DE gels resulted in 1,499 detectedand matched protein spots. All data was pre-processedand analyzed as described in Supplementary data 3.

With an aim to identify a reduced number of protein spotswith predictive value we applied multivariate predictivemodeling to the data. The modeling was carried out in twosteps. First, a training set consistingof 36OND/HCandnine PPSwas used to select the protein spots with the highestpredictivity and to optimize model parameters. In the nextstep the predictivity of the model and the associated selectedprotein spots was evaluated using an independent test setconsisting of six PPS and seven OND/HC samples (Fig. 1).

Fig. 1 –Data analysis strategy. The model optimization wasperformed on a set of 1499 spots that included 36 OND/HCand 9 PPS samples. In the optimization procedure, this set ofsamples was randomly divided into training and test sets,repeatedly, to obtain an estimate of the predictionperformance. From this analysis, the spots were ranked byimportance in the model, and the top 13 spots that wereidentified using mass spectrometry, were chosen toconstitute the finalmodel. The samples used to train the finalmodel, were the 36+9 samples used in the modeloptimization, and this model was tested on a newindependent set of samples, 6 PPS and 7 OND samples.

Please cite this article as: Gonzalez H, et al, Identification of noveImplications for diagnosis, neurodegeneration and neuroinflamm

Additionally, we compared the differentially expressed pro-teins in CSF from PPS with expression levels of same proteinsin CSF from SPMS (17 age-matched patients).

We used the Partial Least Squares (PLS) method [27] to buildpredictive models of the data. To optimize model parametersand to select important protein spots we applied a cross-validation procedure to the training set data, coupled with avariable (corresponding to a protein spot) selection procedure.Briefly, in the cross-validation procedure the training set isinternallydivided into trainingandtest sets randomly, and this isrepeated anumber of times. Theprediction rate is recordedas anaverage of the test set predictions over all cross-validationrounds. Note that the variable (protein spot) selection is onlyperformed on the training set, as inclusion of the test set in thevariable selection can result in substantial overestimation of thepredictive ability of the model [28]. The model optimization wasperformed on two datasets resulting from two different ap-proaches to handling missing values (Supplementary data 3.1).

Finally, a test set evaluation was then performed to assessthe predictivity of the selected model and associated proteinspots using six PPS and seven OND/HC independent samples.

The PLS modeling was performed using Matlab (The Math-works, Inc, Natick, MA, USA) and the PLS toolbox (EigenvectorResearch, Wenatchee, WA, USA). Hierarchical clustering wasperformed using Spotfire (http://www.spotfire.com).

2.5. Protein identification

Protein spots of interest were excised from gels using theEXQUEST spot cutter robot (Bio-Rad, Hercules, CA, USA) andtransferred to 96-well plates. Up to 6 protein spots of the samespot identity were pooled to each well in order to facilitate theidentification of low abundant protein spots. The excised gelplugs were subjected to destaining using a wash solutionconsisting of 70% ACN (LC.MS grade acetonitrile, Fischer Scien-tific, Loughborough, UK) in 25 mM ammonium bicarbonate(Sigma, St. Louis, MO, USA) for 10 min. This washing procedurewas repeated three times. Finally, gel plugs were speed vacuumdried until dryness. Reductionwas performed at 56 °C for 30minby addition of 20 µL of 20 mM DTT solution (Genomic SolutionsInc., Ann Arbor, MI, USA). After cooling of samples to roomtemperature 20 µL of 55 mM iodoacetamide were added andsampleswerealkylated for 45min in thedark. Supernatantswereremoved andACNwas added to shrink the gel plugs. About 15 μLtrypsin (modified/sequencing grade, Promega,Madison,WI,USA)solution (0.075 μg) was added to the gel plugs and digestion wasperformedovernight at 37 °C. Extractionwasperformed for 1hbyadding 20 µL of extraction solution (1% ACN, (Fischer Scientific,UK) 0.1%TFA,Flukapuriss. p.a. forHPLC, Steinheim,Switzerland).

During the peptide preparation two different systems wereused to separate peptides onto MALDI target plates: a micro-technology workstation [29] and a nano-LC on an 1100 NanoflowProteomics Solution system (Agilent Technologies, Böblingen,Germany). (Supplementary data 4.1). Fractions spotted onto theMALDI targets were analyzed in MS and MS/MS mode using the4700 Proteomics Analyzer (Applied Biosystems, Foster City, CA,USA) MALDI-TOF/TOF instrument. All MS/MS data from theMALDI-TOF/TOF instrument were acquired using 4700 Explorersoftware (AppliedBiosystems, FosterCity, CA,USA) allowingnon-redundant and fully automated selectionof precursors forMS/MS


http://expasy.org/swiss-2dpage/viewer

http://www.spotfire.com



ARTICLE IN PRESS

acquisition (Supplementary data 4.2). Resulting MS/MS dataobtained from the MALDI-TOF/TOF instrument were analyzedusing GPS Explorer (Applied Biosystems, Foster City, CA, USA)software, which invoked a MASCOT 2.2.0.3 (Matrix Science,London, UK) database search using the human subset of in-house protein sequence databases (Genseq P, RefSeqP, PDB, PIR,SwissProt and TREMBL and GChuman). Mass tolerance for theprecursor ionwas set to 50 ppm;mass tolerance for the fragmentions was 0.2 Da. Carbamidomethylation was used as fixedmodification and Oxidation (M) was allowed as a variablemodification and mis-cleaveage was set to 1. MASCOT peptidehitswithaMOWSEscoreequalorhigher than35beingconsideredaspositivehits.Aprotein identity comprisedofat least two trypticfragments and/or combined with MALDI spectra scoring.

2.6. ELISA

Determinations of kallikrein 6 levels were assessed in the CSF ofPPS (n=37) andOND (n=30)patientsusinga commercial kallikrein6 ELISA kit (IBEX Pharmaceuticals, Inc. Montreal, Canada) accord-ing to the manufactures instructions. CSF samples were diluted1:50 andkallikrein 6valueswere calculated fromastandard curveand expressed as ng/ml. Differences in CSF Kallikrein 6 levelswere tested for significance using the non-parametric MannWhitney test (GraphPad Prism 3.0, La Jolla, CA, USA).

3. Results

To date, no proteomic study has to our knowledge beenperformed on CSF from PPS patients. In our study there were142 patients and controls in total, in which comparative CSF

Fig. 2 – (A) Prediction of independent patient samples, the test serepresented by the dashed line at 0.425, was determined from themissing values. The number of correctly predicted samples is 12have been obtained by chance (see supplementary data for detay-values of the samples used to build the model, the training se


protein expression data were obtained from 15 PPS patients and60 non-PPS controls. Moreover, protein Kallikrein 6 levels werefurther evaluated in 37 PPS patients and 30 OND patients withELISA.

3.1. Model optimization and variable selection

The PLS models with one component gave the best discrimina-tionbetweenPPSandcontrols for the training set. Thepredictionrate remained relatively constant as variableswere dropped andit was observed that about 5–15 variables were sufficient toachieve ahighprediction rate of PPS (Supplementary data Fig. 1).The average test set prediction rate in the cross-validationprocedure was 90% and 99% correctly predicted samples fromthe twodatasets, respectively.With thePLSmethod it is possibleto determine the importance of a variable (protein spot) for themodel from the VIP parameter (Variable Importance in theProjection), see Supplementary data 3. VIP scores from the twomodels were combined to form an overall variable (protein spot)importance rank and the top 13 protein spots were chosen forthe final model.

3.2. Prediction of independent test samples

To evaluate the model and the associated 13 protein spots weused an independent test set consisting of six PPS and sevenOND/HC samples. The prediction result of the test set ispresented in Fig. 2. Six-out-of-seven OND/HC samples werecorrectly predicted and all six PPS samples were correctlypredicted. Thus the number of correctly predicted samples is12/13 (92.3%), with a p-value of 0.0017 that these results wouldhave been obtained by chance (Supplementary data 3.4).

t, using the 13-variable prediction model. The cut off-value,twomodels originated from the two approaches to handling

of 13 (92.3%), with a p-value of 0.0017 that these results wouldils regarding the statistical test). For reference, predictedt, are shown in (B).



Table 2 – Identification of the top-13 most predictive protein spots excised from 2DE gels

Spotnumber(SSP)

Protein identity ⁎ indicatesprotein fragment

Acc.# Mascotscore

Matchedpeptides

Coverage(%)

TheoreticalMW (kDa)

TheoreticalpI

ObservedMW (kDa)

ObservedpI (gel)

Firstaminoacid

Sequence

3403 Gelsolin⁎ P06396 166 4 5 80,876 5.58 54 5.08 503 EGGQTAPASTR534 AGALNSNDAFVLK679 YIETDPANRDR690 RTPITVVK

3408 Gelsolin⁎ P06396 192 4 5 80,876 5.58 54 5.14 503 EGGQTAPASTR534 AGALNSNDAFVLK679 YIETDPANRDR690 RTPITVVK

4414 Peptidylglycine alpha-AmidatingMonooxygenase⁎(PAM)

P19021 186 3 3 97,051 6.00 51 5.40 416 NNLVIFHR504 EGPVLILGR549 IVQFSPSGK

5617 Hemopexin P02790 235 5 9 52,254 6.57 77 5.75 91 NFPSPVDAAFR102 QGHNSVFLIK197 YYCFQGNQFLR208 FDPVRGEVPPR213 GEVPPRYPR

5631 Hemopexin P02790 259 4 7 52,254 6.57 77 5.65 91 NFPSPVDAAFR197 YYCFQGNQFLR208 FDPVRGEVPPR213 GEVPPRYPR

5632 Hemopexin P02790 195 4 7 52,254 6.57 77 5.56 91 NFPSPVDAAFR197 YYCFQGNQFLR208 FDPVRGEVPPR213 GEVPPRYPR



7208 Hemopexin ⁎ P02790 34 1 2 52,254 6.57 34 6.20 332 GGYTLVSGYPK7310 Glutathione P48637 54 2 3 52,523 5.67 38 6.16 26 ALAEGVLLR

Synthetase(GSHB) ⁎ 222 AIENELLAR7328 Hemopexin ⁎ P02790 45 1 2 52,254 6.57 34 6.05 332 GGYTLVSGYPK8206 Hemopexin ⁎ P02790 46 1 2 52,254 6.57 28 6.37 332 GGYTLVSGYPK8221 Kallikrein 6 Q92876 138 3 10 27,523 7.15 28 6.31 79 QRESSQEQSSVVR

81 ESSQEQSSVVR118 LSELIQPLPLER

The SSP spot numbers and observed Mr/pI refer to the 2DE gel positions shown in Fig. 2. Identification was performed by mass spectrometry as described in Section 2.5. 5JO

UR

NA

LO

FPR

OT

EO

MIC

SX

X(2009)

XX

X–X

XX

ARTICLE

INPRESS

Pleasecite

this

articleas:G

onzalez

H,et

al,Identificationofn

ovelcandidate

proteinbiom

arkersforth

epost-polio

syndrom

e—

Implication

sfor

diagnosis,n

eurodegen

erationand

neu

roinflam

mation

,JProt(2009),doi:10.1016/j.jprot.2008.11.014


Fig. 3 –2-DE separation of CSF proteins from one PPS patient. The positions of the top-13 identified protein spots are shown,hemopexin, gelsolin, kallikrein-6, peptidylglycine alpha-hydroxylating monooxygenase (PAM) and glutathione synthetase(GSHB). For reference, the positions of serotransferrin, alpha-1-antitrypsin and apo-lipoprotein A1 are indicated. The right panelshow zoom-in areas of identified proteins and SSP number. Mass spectrometry data for the top-13 proteins are shown inTable 2.


ARTICLE IN PRESS

3.3. Identity and expression levels of proteins

Proteinspotsof interestwere cut out from2D-gels and identifiedby mass spectrometry (Table 2). A representative 2D-gel isdepicted in Fig. 3 and the locations of the top ranked 13 proteinspots are magnified. Each spot is given a unique database SSPnumber by the PDQuest™ software. Table 2 and Fig. 3 presentprotein identities obtained by mass spectrometry analysestogether with theoretical and observed molecular weight/isoelectric points (Mr/pI). The identitieswere gelsolin fragments(SSP3403and3408), full-lengthhemopexin (SSP5617, 5631, 5632,6605 and 6611), hemopexin fragments (SSP 7208, 7328 and 8206),kallikrein 6 (SSP 8221), peptidylglycine alpha-amidating mono-oxygenase (PAM, SSP 4414) and glutathione synthetase (GSHB,SSP 7310), respectively.

The relative importance of the protein spots for the predic-tion model was determined by calculating the PLS VIP scores,both for the training set and for a model consisting of allsamples, i.e. the combined training and test sets The combinedmodel was constructed so as to use as much of the informationavailable as possible in determining protein spot importance(Supplementary data Table 1). Note that 11 of the top rankedprotein spots represent only three different protein entities,gelsolin, kallikrein 6 and hemopexin, respectively. This indi-cates per se that these proteins may have a key role in PPS.Expression levels and descriptive statistics for the top-13proteins are presented in Table 2a Supplementary data. Toprovidea summaryoverviewof theexpression levels of these 11


protein spots, three meta-variables were derived based on thesum of spots with the same identity corresponding to theexpression level for the proteins. Fig. 4 reveals that gelsolinfragments, hemopexin full-length and fragments and kallik-rein-6 are all differentially expressed, not only compared to thecontrol group but also compared to the SPMS group. (Expressionlevels anddescriptive statistics are listed inSupplementarydataTable 2a and b for protein spots and the derivedmeta-variables,respectively).

We observed a 2.4-fold increase of two protein spotsrepresenting gelsolin fragments (Fig. 4 and (Supplementarydata Table 2b), while no significant change in the level of full-length gelsolin was observed (data not included). This mostlikely reflects an increased enzymatic cleavage of the proteinandpossibly, inparallel, an increased compensatory productionof gelsolin. Gelsolin cleavage is tightly associated with apopto-sis, and both pro-apoptotic and anti-apoptotic roles have beendescribed [30].

There was a 2.3-fold decrease of full-length hemopexinvariants in parallel with a 4.2-fold increase of fragmentedhemopexin, indicating a proteolytic cleavage of the protein(Fig. 4) and (Supplementary data Table 2b). Interestingly,this is in contrast to the classical role of hemopexin whichbelongs to the family of acute-phase proteins that areusually induced during inflammatory events [31].

The kallikrein 6 protein spot intensity increased 5.3-fold(Fig. 4 and Supplementary data Table 2a) and is likely to be afull-length variant (according to observed versus theoretical



Fig. 4 –Box-plots showing expression levels in ppm for 43 OND/HC, 15 PPS and 17 MS-SP samples. The boxes are defined bythe 25th and 75th percentile, thus 50% of the data is represented within each box. The lower/upper whiskers indicate thesmallest/largest value greater/less than or equal to the first/third quartile minus/plus 1.5*interquartile range. The median isrepresented by a left-arrow and the mean is represented by a right-arrow.


ARTICLE IN PRESS

Please cite this article as: Gonzalez H, et al, Identification of novel candidate protein biomarkers for the post-polio syndrome—Implications for diagnosis, neurodegeneration and neuroinflammation, J Prot (2009), doi:10.1016/j.jprot.2008.11.014


Fig. 5 –Result from the ELISA on Kallikrein 6 levels in PPS(N=37) and OND (N=30) patients. Mann–Whitney test wasused to test significance, p value=0.0007. The mean value isrepresented with a line, 710 ng/ml and 590 ng/ml for PPS andOND groups, respectively.


ARTICLE IN PRESS

molecular weight) of the protein (Table 2). Kallikrein 6 is aserine protease and is preferentially expressed in the CNS athighest levels in the spinal cord and brainstem [32], and hasa trypsin-like broad substrate enzymatic activity [33,34]. Inaddition to this single kallikrein 6 spot (SSP 8221) weobserved one other spot (SSP 8209) identified as kallikrein6 (data not included), although at lower VIP scores (VIP rank35). Interestingly both kallikrein 6 spots exhibited increasedlevels in PPS, which support a possible role of kallikrein 6 inPPS. Moreover, the expression levels in the SPMS group forthe proteins are at the same level as for the OND/HC-group

Fig. 6 –Hierarchical clustering of all samples. On the x-axis and himodel on the y-axis. Hierarchical clustering is an unsupervisedsimilarity with respect to expression pattern. The hierarchical cland PPS. Samples indicated by “X” represent the new independewithin the PPS group (sample 23). The table to the right show prgelsolin fragments, kallikrein 6, hemopexin, hemopexin fragmemonooxygenase, and GSHB_HUMAN equals glutathione synthet


(Fig. 4). The increased expression level of kallikrein 6 wasconfirmed using a commercial ELISA (methods) with a pvalue of 0.0007 (Fig. 5). It is not surprising to observe adifference in the fold change value of kallikrein 6 expressionlevels between the two methods since proteomics measurespecific forms of the protein whereas ELISA assesses allforms recognized by the antibody.

Two additional proteins of possible relevance to PPS wereidentified: peptidylglycine alpha-amidating monooxygenase(PAM), with 3.2-fold decrease, which plays a key role in thebiosynthesis of many neuronal peptides and is present inneurons as well as in oligodendrocytes [35]. It has beenspeculated that an increase of PAM activity in CSF frompatients with MSmay arise from demyelination and oligoden-drocyte destruction [36]. Glutathione synthetase (GSHB), with1.9-fold increase, is a key enzyme in the biosynthesis ofglutathione. Glutathione is important for a variety of biologicalfunctions including protection of cells from oxidative damageby free radicals, detoxification of xenobiotics and membranetransport [37]. However, these two proteins had the lowest VIPscores and are thus of least significance for the predictivemodel.

To provide an overview of all samples in relation to theexpression levels of the protein spots used in the modelbuilding, a two-dimensional hierarchical clustering wasconducted (Fig. 6). Hierarchical clustering is an unsuper-vised method which allows data to self-organize accordingto similarity with respect to expression pattern. A clearseparation of the two patient groups PPS and OND/HC wasobserved, and all samples except one cluster within thecorrect diagnostic group. This is the same sample that

erarchical clustering of the top-13 spots used in the predictionmethod, which allows data to self-organize according toustering shows good separation of the two groups, OND/HCnt test samples. Only one of the OND samples sample clusterotein identities and corresponding SSP spot number are:nts, PAM equals peptidylglycine alpha-amidatingase.




ARTICLE IN PRESS

was incorrectly classified in the PLS model (Fig. 2a). Inaddition, one can observe that proteins of same identitytend to cluster together, e.g. the two gelsolin fragmentsand the five hemopexin spots representing the full-lengthprotein.

4. Discussion

Protein analysis employing classical proteomics combinedwith multivariate modeling and identification using massspectrometry resulted in the discovery of three differen-tially expressed proteins or their fragments in PPS samplesas compared to in controls. This firstly suggests that theseproteins may exert key roles in PPS patophysiology.Secondly, these proteins and their fragments representpotential candidates as biomarkers for the disease. Tomerit as true biomarkers studies will be required in largermaterials of PPS and a variety of other CNS diseases.Notably, however, in comparison with samples from SPMS(being an age-matched control group with ongoing inflam-mation and neuronal destruction), the most predictiveproteins were specific for PPS. In view of the proteinexpression observed herein in PPS and the functional roleof these proteins and potential reasons for their fragmen-tations, as discussed below, the data strongly support theoccurrence of ongoing exaggerated nervous tissue damagein PPS, the underlying cause of which still remains unclear.However, previous observations of the increased expressionof pro-inflammatory cytokines in the CSF and their down-modulation upon IVIG treatment combined with a bene-ficial effect on clinical measures of the disease [24,38,39],strongly argues for a chronic inflammation causing thenervous tissue damage. The underlying cause for such achronic insidious inflammation decades after the acuteparalytic disease still remains elusive. Although there is nounequivocal demonstration of latent poliovirus in PPS, thispossibility cannot be excluded. Another possible explana-tion could be an aberrant, perhaps autoimmune responseto nervous tissue, once and originally triggered by theinfection.

The increase of kallikrein 6 levels can be regarded asbeing part of a CNS innate immunity response to a varietyof insults, both traumatic and primary inflammatory events[40]. Kallikrein 6 is normally expressed in neurons andoligodendrocytes, and is strongly up-regulated in astrocytesand macrophages/microglia during CNS pathologies such asspinal cord trauma and in human or experimental inflam-matory demyelinating diseases [40–42]. It has also beendetected in the CSF and CNS of neurodegenerative diseasessuch as Alzheimer's and Parkinson's disease [43]. Ourfinding of an increase in kallikrein 6 in the CSF of personswith PPS provides a strong additional argument for anongoing nervous tissue damaging process underlying thesyndrome. In view of its induction in lesions of primaryCNS inflammatory diseases [42,44], we speculate that pro-inflammatory cytokines such as TNF-α and IFN-γ may alsobe considered to be involved, in particular in view of theirknown increased production in PPS [24,38,45]. As with mostinflammatory mediator and effector molecules, both bene-


ficial and detrimental effects can be postulated for kallik-rein 6. On the potential beneficial side, the enzymemodifies the conditions for neurite outgrowth after spinalcord trauma [40]. On the detrimental side, the enzymemediates toxicity against myelin/oligodendrocytes inexperimental autoimmune encephalomyelitis [42]. Thuswhile the exact consequences of the observed increasedKallikrein 6 levels in PPS are as yet unclear, it may play acentral role in the late symptomatology subsequent topolio.

In the present study fragments of gelsolin were themost predictive among the proteins present in the CSF ofpersons with PPS. It has been reported that caspase 3cleaves gelsolin in the pro-apoptotic pathway [46]. Further-more, TNF-α induces caspase 3-mediated gelsolin cleavagein muscle cells [47]. In its anti-apoptotic role, gelsolin canprevent apoptosis via the mitochondrial pathway [48]. Ourdata suggest that gelsolin is cleaved by caspase-3 in PPSfor the following reasons: (I) the cleavage site of caspase 3produces an acidic protein variant which matches wellwith observed Mr/pI, and (II) all four identified trypticpeptides are located within the same acidic protein(Table 2). Hence the detected fragmentation of gelsolinsuggests increased caspase 3 activity during PPS andconsequently an increased intrathecal apoptotic activity.

We observed a decrease in levels of the full-lengthhemopexin protein, along with increased levels of severalhemopexin-fragments in CSF. Due to the well-known factthat hemopexin migrates as multiple glycosylated variantsin a 2D-gel separation [49], we conclude that the observedparallel decrease of five full-length hemopexin variantsrepresents highly reliable data. The most well character-ized function of hemopexin is to bind heme and to protectagainst oxidative stress through receptor-mediated induc-tion of heme oxygenase [50]. Interestingly, the role ofhemopexin seems to be different in CNS as compared toits systemic functions. It has been reported that hemo-pexin can be induced in response to tissue damage in theCNS [51]. Changes in hemopexin expression patterns havebeen recorded in an inflammatory kidney disease, Mini-mal Change Disease (MCD), and plasma samples fromsuch persons exhibited proteolytic activity resulting indegradation of extracellular matrix [52]. Our data areconsistent with an increased proteolytic activity affectinghemopexin in the CNS of persons with PPS, perhaps beinga secondary event to chronic inflammation. However, thedetailed reason for the fragmentation of hemopexin in PPSis unknown and remains to be further investigated. Itshould be noted that full-length gelsolin and hemopexinhave previously been proposed as biomarker candidates inneurodegenerative diseases [53,54] and that our observa-tions focus on the fragmentation of these proteins. Tofurther evaluate hemopexin and gelsolin as biomarkerswe therefore suggest western blot analysis rather thanELISA.

CSF mononuclear mRNA expression of TNF-α and IFN-γ,being two key pro-inflammatory cytokines, have beenassessed in all individuals studied presently in a previousstudy and all had elevated levels [24,38,45]. An increase ofCSF TNF-α protein was recently confirmed in the CSF of




ARTICLE IN PRESS

PPS patients [18]. This suggests that pro-inflammatorycytokines may be key players in the development andmaintenance of PPS. The role of an inflammatory processis further supported by previous findings of inflammatoryinfiltrates in the spinal cord of post-mortem PPS material[55–57], as well as through signs of inflammation inaffected muscles [7,58].

Since we have not had access to CSF from persons withprior poliomyelitis, but with no signs of increasing latesymptomatology as in PPS, we cannot know if the ob-served aberrations are characteristic for prior poliomyelitisoverall, or characteristic of the late deterioration. Studiesof persons with stable post-polio disabilities are thereforewarranted. However, we hypothesize that the inflamma-tory changes and protein aberrations could be pathogeni-cally relevant even if they would be present in priorpoliomyelitis without apparent late squealae. Differentindividuals could well be differentially sensitive forchronic low-grade CNS inflammation until critical thresh-olds are reached, resulting in the varieties of late sequelae.Furthermore, there is experimental evidence in the ratspecies for genetically determined differences in thesusceptibility of motor neurons to non-specific andinflammatory damage, as rat non-MHC genes drasticallyaffect nerve avulsion-induced motor neuron degenerationand CNS innate immunity [59–61]. If there are similargenes in action in the human species, the outcome in theform of neuronal damage may well vary between indivi-duals at similar levels of chronic inflammation as areapparent late after paralytic poliomyelitis.

In conclusion, we herein demonstrate a protein profile,based on its high predictive value, has the potential toserve as a diagnostic biomarker for PPS. The proteinsidentified in this study are known to be involved indifferent pathways associated with tissue damage andapoptosis. These data and previous observations ofinflammation and cytokine production provide strongsupport for the hypothesis that PPS is caused by an activeinflammatory and neurodegenerative process. There isconsequently potential for various modes of anti-inflam-matory and/or neuroprotective therapy.

Role of the funding source

The sponsor, AstraZeneca R&D, Sweden, did not interferewith the study design or interpretation of the data nor inthe decision to submit the paper for publication. Thecorresponding author, as well as all other authors, havefull access to all the data in the study and have finalresponsibility for the decision to submit for publication.

Conflict of interest

There are no other known conflicts of interest besides apatent application (reference number US 60/944,816)entitled “Diagnostic Method”. This application was filed2007-06-19 in US.


Acknowledgements

The proteomics have been done at AstraZeneca R&D, theclinical parts, and sample collections have been supportedby grants from the Gustaf V foundation and the Swedishmedical research council.

The Swedish research council, Gustav V foundation.Hugh Salter and Jenny Flygare for reviewing the manu-

script. Jon D Randall for technical assistance and supportregarding the PDQuest software. Lisbet Broman for herexcellent coordination of the clinical data.

Appendix A. Supplementary data

Supplementary data associated with this article can be found,in the online version, at doi:10.1016/j.jprot.2008.11.014.

R E F E R E N C E S

[1] Trojan DA, Cashman NR. Post-poliomyelitis syndrome.Muscle Nerve 2005;31(1):6–19.

[2] Farbu E, Gilhus NE, Barnes MP, Borg K, de Visser M, Driessen A,et al. EFNS guideline on diagnosis and management ofpost-polio syndrome. Report of an EFNS task force. Eur JNeurol 2006;13(8):795–801.

[3] McComas AJ, Quartly C, Griggs RC. Early and late losses ofmotor units after poliomyelitis. Brain 1997;120(Pt 8):1415–21.

[4] Grimby G, Stålberg E, Sandberg A, Sunnerhagen KS. An 8-yearlongitudinal study of muscle strength, muscle fiber size, anddynamic electromyogram in individuals with late polio.Muscle Nerve 1998;21(11):1428–37.

[5] Bartholdi D, Gonzalez H, Borg K,Melki J. Absence of SMN genedeletion in post-polio syndrome. Neuromuscul Disord2000;10(2):99.

[6] Cashman NR, Trojan DA. Correlation of electrophysiologywith pathology, pathogenesis, and anticholinesterasetherapy in post-polio syndrome. Ann N Y Acad Sci1995;753:138–50.

[7] Dalakas M, Illa I. Post-polio syndrome: concepts in clinicaldiagnosis, pathogenesis, and etiology. Adv Neurol1991;56:495–511.

[8] Dalakas MC. Pathogenetic mechanisms of post-poliosyndrome: morphological, electrophysiological, virological,and immunological correlations. Ann N Y Acad Sci1995;753:167–85.

[9] Dalakas MC. Pro-inflammatory cytokines and motor neurondysfunction: is there a connection in post-polio syndrome?J Neurol Sci 2002;205(1):5–8.

[10] Dalakas MC, Sever JL, Madden DL, Papadopoulos NM,Shekarchi IC, Albrecht P, et al. Late postpoliomyelitismuscular atrophy: clinical, virologic, and immunologicstudies. Rev Infect Dis 1984;6(Suppl 2):S562–7.

[11] Dinsmore S, Dambrosia J, Dalakas MC. A double-blind,placebo-controlled trial of high-dose prednisone for thetreatment of post-poliomyelitis syndrome. Ann N Y Acad Sci1995;753:303–13.

[12] Halstead LS. Post-polio syndrome. Sci Am 1998;278(4):42–7.[13] Leon-Monzon ME, Illa I, Dalakas MC. Expression of poliovirus

receptor in human spinal cord and muscle. Ann N Y Acad Sci1995;753:48–57.

[14] Pelletier I, Duncan G, Pavio N, Colbére-Garapin F. Molecularmechanisms of poliovirus persistence: key role of capsid





ARTICLE IN PRESS

determinants during the establishment phase. Cell Mol LifeSci 1998;54(12):1385–402.

[15] Rekand T, et al. Fcgamma receptor IIIA polymorphism as arisk factor for acute poliomyelitis. J Infect Dis 2002;186(12):1840–3.

[16] Roivainen M, Kinnunen E, Hovi T. Twenty-one patientswith strictly defined postpoliomyelitis syndrome: nopoliovirus-specific IgM antibodies in the cerebrospinal fluid.Ann Neurol 1994;36(1):115–6.

[17] Sharief MK, Hentges R, Ciardi M. Intrathecal immuneresponse in patients with the post-polio syndrome. N Engl JMed 1991;325(11):749–55.

[18] Farbu E, Rekand T, Vik-Mo E, Lygren H, Gilhus NE, Aarli JA.Post-polio syndrome patients treated with intravenousimmunoglobulin: a double-blinded randomized controlledpilot study. Eur J Neurol 2007;14(1):60–5.

[19] Fordyce CB, Gagne D, Jalili F, Alatab S, Arnold DL,Da Costa D, et al. Elevated serum inflammatory markersin post-poliomyelitis syndrome. J Neurol Sci2008;271(1–2):80–6.

[20] Zerr I, et al. Diagnosis of Creutzfeldt–Jakob disease bytwo-dimensional gel electrophoresis of cerebrospinal fluid.Lancet 1996;348(9031):846–9.

[21] Puchades M, Hansson SF, Nilsson CL, Andreasen N, BlennowK, Davidsson P. Proteomic studies of potentialcerebrospinal fluid protein markers for Alzheimer's disease.Brain Res Mol Brain Res 2003;118(1–2):140–6.

[22] Choe LH, et al. Studies of potential cerebrospinal fluidmolecular markers for Alzheimer's disease. Electrophoresis2002;23(14):2247–51.

[23] Halstead LS, Rossi CD. Post-polio syndrome: clinicalexperience with 132 consecutive outpatients. Birth DefectsOrig Artic Ser 1987;23(4):13–26.

[24] Gonzalez H, Khademi M, Andersson M, Piehl F, Wallström E,Borg K, et al. Prior poliomyelitis-IvIg treatment reducesproinflammatory cytokine production. J Neuroimmunol2004;150(1–2):139–44.

[25] Newcombe J, Eriksson B, Ottervald J, Yang Y, Franzén B.Extraction and proteomic analysis of proteins from normaland multiple sclerosis postmortem brain. J Chromatogr BAnalyt Technol Biomed Life Sci 2005;815(1–2):191–202.

[26] Berggren K, Chernokalskaya E, Steinberg TH, Kemper C, LopezMF, Diwu Z, et al. Background-free, high sensitivity staining ofproteins in one- and two-dimensional sodium dodecylsulfate-polyacrylamide gels using a luminescent rutheniumcomplex. Electrophoresis 2000;21(12):2509–21.

[27] Wold S, Sjostrom M, Eriksson L. PLS-regression: a basic toolof chemometrics. Chemometr Intell Lab Syst 2001;58:109–30.

[28] Ambroise C, McLachlan GJ. Selection bias in gene extractionon the basis of microarray gene-expression data. Proc NatlAcad Sci U S A 2002;99(10):6562–6.

[29] Wallman L, Ekström S, Marko-Varga G, Laurell T, Nilsson J.Autonomous protein sample processing on-chip usingsolid-phase microextraction, capillary force pumping, andmicrodispensing. Electrophoresis 2004;25(21–22):3778–87.

[30] Leifeld L, et al. Anti-apoptotic function of gelsolin in fasantibody-induced liver failure in vivo. Am J Pathol 2006;168(3):778–85.

[31] Baumann H, Gauldie J. The acute phase response. ImmunolToday 1994;15(2):74–80.

[32] Scarisbrick IA, Towner MD, Isackson PJ. Nervoussystem-specific expression of a novel serine protease:regulation in the adult rat spinal cord by excitotoxic injury.J Neurosci 1997;17(21):8156–68.

[33] Bernett MJ, Blaber SI, Scarisbrink IA, Dhanarajan P, ThompsonSM, Blaber M. Crystal structure and biochemicalcharacterization of human kallikrein 6 reveals that atrypsin-like kallikrein is expressed in the central nervoussystem. J Biol Chem 2002;277(27):24562–70.


[34] Blaber SI, et al. Enzymatic properties of ratmyelencephalon-specific protease. Biochemistry2002;41(4):1165–73.

[35] Bolkenius FN, Ganzhorn AJ. Peptidylglycine alpha-amidatingmono-oxygenase: neuropeptide amidation as a target fordrug design. Gen Pharmacol 1998;31(5):655–9.

[36] Tsukamoto T, Noguchi M, Kayama H, Watanabe T, Asoh T,Yamamoto T. Increased peptidylglycine alpha-amidatingmonooxygenase activity in cerebrospinal fluid of patientswith multiple sclerosis. Intern Med 1995;34(4):229–32.

[37] Meister A, Anderson ME. Glutathione. Annu Rev Biochem1983;52:711–60.

[38] Gonzalez H, Khademi M, Andersson M, Wallström E, Borg K,Olsson T. Prior poliomyelitis-evidence of cytokine productionin the central nervous system. J Neurol Sci 2002;205(1):9–13.

[39] Gonzalez H, Stibrant Sunnerhagen K, Sjöberg I, Kaponides G,Olsson T, Borg K. Intravenous immunoglobulin for post-poliosyndrome: a randomised controlled trial. Lancet Neurol2006;5(6):493–500.

[40] Scarisbrick IA, Sabharwal P, Cruz H, Larsen N, VAndell AG,Blaber SI, et al. Dynamic role of kallikrein 6 in traumatic spinalcord injury. Eur J Neurosci 2006;24(5):1457–69.

[41] Scarisbrick IA, Blaber SI, Luccinetti CF, Genain CP, Blaber M,Rodriguez M. Activity of a newly identified serine proteasein CNS demyelination. Brain 2002;125(Pt 6): 1283–96.

[42] Blaber SI, Ciric B, Christophi GP, BernettMJ, BlaberM, RodriguezM, et al. Targeting kallikrein 6 proteolysis attenuates CNSinflammatory disease. Faseb J 2004;18(7):920–2.

[43] Diamandis EP, Yousef GM, Petraki C, Soosaipillai AR. Humankallikrein 6 as a biomarker of alzheimer's disease. ClinBiochem 2000;33(8):663–7.

[44] Scarisbrick IA, Blaber S, Tingling J, Rodriguez M, Blaber M,Christophi G. Potential scope of action of tissue kallikreinsin CNS immune-mediated disease. J Neuroimmunol2006;178(1–2):167–76.

[45] Mitka M. Aging brings new challenges for polio survivors.Jama 2006;296(14):1718–9.

[46] Kothakota S, Azuma T, Reinhard C, Klippel A, Tang J,Chu K, et al. Caspase-3-generated fragment of gelsolin:effector of morphological change in apoptosis. Science1997;278(5336):294–8.

[47] Geng YJ, Azuma T, Tang JX, hartwig JH, Muszynski M, WuQ, et al. Caspase-3-induced gelsolin fragmentationcontributes to actin cytoskeletal collapse, nucleolysis,and apoptosis of vascular smooth muscle cells exposedto proinflammatory cytokines. Eur J Cell Biol 1998;77(4):294–302.

[48] Kusano H, Shimizu S, Koya RC, Fujita H, Kamada S, KuzumakiN, et al. Human gelsolin prevents apoptosis by inhibitingapoptoticmitochondrial changes via closing VDAC. Oncogene2000;19(42):4807–14.

[49] Ogata Y, Charlesworth MC, Muddiman DC. Evaluation ofprotein depletion methods for the analysis oftotal-, phospho- and glycoproteins in lumbar cerebrospinalfluid. J Proteome Res 2005;4(3):837–45.

[50] Tolosano E, Altruda F. Hemopexin: structure, function, andregulation. DNA Cell Biol 2002;21(4):297–306.

[51] Madore N, Camborieux L, Bertrand N, Swerts JP. Regulation ofhemopexin synthesis in degenerating and regenerating ratsciatic nerve. J Neurochem 1999;72(2):708–15.

[52] Bakker WW, van Dael CM, Pierik LJ, van Wijk JA, Nauta J,Borghuis T, et al. Altered activity of plasma hemopexin inpatients with minimal change disease in relapse. PediatrNephrol 2005;20(10):1410–5.

[53] Castano EM, Roher AE, Esh CL, Kokjohn TA, Beach T.Comparative proteomics of cerebrospinal fluid inneuropathologically-confirmed Alzheimer's diseaseand non-demented elderly subjects. Neurol Res2006;28(2): 155–63.




ARTICLE IN PRESS

[54] Qiao H, Koya R, Nakagawa K, Tanaka H, Fujita H, Takimoto M,et al. Inhibition of Alzheimer's amyloid-beta peptide-inducedreduction of mitochondrial membrane potential andneurotoxicity by gelsolin. Neurobiol Aging 2005;26(6):849–55.

[55] Pezeshkpour GH, Dalakas MC. Pathology of spinal cord inpost-poliomyelitis muscular atrophy. Birth Defects Orig ArticSer 1987;23(4):229–36.

[56] Pezeshkpour GH, Dalakas MC. Long-term changes inthe spinal cords of patients with old poliomyelitis.Signs of continuous disease activity. Arch Neurol 1988;45(5):505–8.

[57] Ikemoto A, Hirano A. Immunohistochemical studies onclustered pericapillary bodies in the spinal cord ofpost-poliomyelitis patients. Acta Neuropathol (Berl) 1996;92(2):164–9.

[58] Dalakas MC. Morphologic changes in the muscles of patientswith postpoliomyelitis neuromuscular symptoms. Neurology1988;38(1):99–104.


[59] Lundberg C, Lidman O, Holmdahl R, Olsson T, Piehl F.Neurodegeneration and glial activation patterns aftermechanical nerve injury are differentially regulated bynon-MHC genes in congenic inbred rat strains.J Comp Neurol 2001;431(1):75–87.

[60] Lidman O, Swanberg M, Horvath L, Broman KW, Olsson T,Piehl F. Discrete gene loci regulateneurodegeneration, lymphocyte infiltration, and majorhistocompatibility complex class II expression in the CNS.J Neurosci 2003;23(30):9817–23.

[61] Swanberg M, Lidman O, Padyukov L, Eriksson P, Åkesson E,Jagodic M, et al. MHC2TA is associated with differential MHCmolecule expression and susceptibility to rheumatoidarthritis, multiple sclerosis and myocardial infarction. NatGenet 2005;37(5):486–94.



Documents

Identification of novel candidate protein biomarkers for ...bme.lth.se/fileadmin/user_upload/Publications/09_JProt_Gonzalez.pdf · Identification of novel candidate protein biomarkers