Multiple sclerosis: Identification and clinical evaluation of novel CSF biomarkers.

Abstract

Multiple sclerosis (MS) is a neuro-inflammatory and neurodegenerative disease that results in damage to myelin sheaths and axons in the central nervous system and which preferentially affects young adults. We performed a proteomics-based biomarker discovery study in which cerebrospinal fluid (CSF) from MS and control individuals was analyzed (n=112). Ten candidate biomarkers were selected for evaluation by quantitative immunoassay using an independent cohort of MS and control subjects (n=209). In relapsing-remitting MS (RRMS) patients there were significant increases in the CSF levels of alpha-1 antichymotrypsin (A1AC), alpha-1 macroglobulin (A2MG) and fibulin 1 as compared to control subjects. In secondary progressive MS (SPMS) four additional proteins (contactin 1, fetuin A, vitamin D binding protein and angiotensinogen (ANGT)) were increased as compared to control subjects. In particular, ANGT was increased 3-fold in SPMS, indicating a potential as biomarker of disease progression in MS. In PPMS, A1AC and A2MG exhibit significantly higher CSF levels than controls, with a trend of increase for ANGT. Classification models based on the biomarker panel could identify 70% of the RRMS and 80% of the SPMS patients correctly. Further evaluation was conducted in a pilot study of CSF from RRMS patients (n=36), before and after treatment with natalizumab.

Full text

Page 1

Multiple sclerosis: Identification and clinical evaluation of
novel CSF biomarkers
Jan Ottervalda,b,⁎, Bo Franzénb, Kerstin Nilssonc, Lars I. Anderssonc, Mohsen Khademia,
Bodil Erikssonb,1, Sven Kjellströmd, György Marko-Vargad,e, Ákos Végvárie,
Robert A. Harrisa, Thomas Laurelle, Tasso Miliotisf, Darius Matuseviciusg, Hugh Salterc,i,
Mats Fermh, Tomas Olssona
aNeuroimmunology Unit, Department of Clinical Neuroscience and Center for Molecular Medicine,
Karolinska Institute at Karolinska University Hospital, Solna, Sweden
bMolecular Pharmacology, AstraZeneca R&D, Södertälje, Sweden
cDisease Biology, AstraZeneca R&D, Södertälje, Sweden
dBiological Sciences, Local Discovery, AstraZeneca R&D Lund, Sweden
eClinical Protein Science & Imaging Group, Biomedical Center Dept. of Measurement Technology and Industrial Electrical Engineering,
Lund University, Lund, Sweden
fTranslational Science, Biosciences, AstraZeneca R&D Mölndal, Sweden
gEarly Development, AstraZeneca R&D, Södertälje, Sweden
hGlobal Discovery, AstraZeneca R&D, Södertälje, Sweden
iLinnaeus Centre for Bioinformatics, Uppsala University, Uppsala, Sweden
A R T I C L E I N F OA B S T R A C T
Multiple sclerosis (MS) is a neuro-inflammatory and neurodegenerative disease that results
in damage to myelin sheaths and axons in the central nervous system and which
preferentially affects young adults. We performed a proteomics-based biomarker discovery
study in which cerebrospinal fluid (CSF) from MS and control individuals was analyzed
(n=112). Ten candidate biomarkers were selected for evaluation by quantitative
immunoassay using an independent cohort of MS and control subjects (n=209). In
relapsing–remitting MS (RRMS) patients there were significant increases in the CSF levels
of alpha-1 antichymotrypsin (A1AC), alpha-1 macroglobulin (A2MG) and fibulin 1 as
compared to control subjects. In secondary progressive MS (SPMS) four additional proteins
(contactin 1, fetuin A, vitamin D binding protein and angiotensinogen (ANGT)) were
increased as compared to control subjects. In particular, ANGT was increased 3-fold in
SPMS, indicating a potential as biomarker of disease progression in MS. In PPMS, A1AC and
A2MG exhibit significantly higher CSF levels than controls, with a trend of increase for
ANGT. Classification models based on the biomarker panel could identify 70% of the RRMS
Keywords:
Multiple sclerosis
CSF
Proteomics
Biomarkers
Prognosis
Treatment
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2
Abbreviations: CIS, clinically isolated syndrome; RRMS, relapsing remitting multiple sclerosis; SPMS, secondary progressive multiple
sclerosis; PPMS, primary progressive multiple sclerosis; CSF, cerebrospinal fluid; HC, healthy controls; OND, other neurological disease;
PLS, Partial Least Squares; VIP, variable importance in the projection; Mr/pI, molecular weight/isoelectric point.
⁎ Corresponding author. Neuroimmunology Unit, Department of Clinical Neuroscience and Centre for Molecular Medicine, CMM, L8:04,
Karolinska Institute at Karolinska Hospital, 171 76, Stockholm, Sweden. Tel.: +46 8 51776246, +46 73 6459165; fax: +46 8 51776248.
E-mail addresses: jan.ottervald@ki.se, jan.Ottervald@astrazeneca.com (J. Ottervald).
1Current address: Recipharm, Södertälje, Sweden.
1874-3919/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.jprot.2010.01.004
available at www.sciencedirect.com
www.elsevier.com/locate/jprot

Page 2

and 80% of the SPMS patients correctly. Further evaluation was conducted in a pilot study of
CSF from RRMS patients (n=36), before and after treatment with natalizumab.
© 2010 Elsevier B.V. All rights reserved.
1.Introduction
Multiple sclerosis (MS) is an inflammatory disease in which
the myelin sheaths and axons are destroyed, with character-
istic focal plaques in the white matter of the central nervous
system. Disease outcome is highly variable between affected
individuals and there is a lack of prognostic markers. For
an individual patient the severity, outcome and the rate of
progression of MS are impossible to predict. Current diagnosis
relies on clinical examination supported by laboratory inves-
tigations including Magnetic Resonance Imaging (MRI) to
visualise lesions, and cerebrospinal fluid (CSF) biochemistry
measurements that include assessment of the oligoclonal
band(s) of IgG and barrier index [1]. To diagnose patients
correctly, demonstration of distribution of lesions in both time
and space is necessary, delaying the time for appropriate
diagnosis [2]. A test measuring biochemical biomarkers could
therefore be a useful tool in the diagnostic process. At present
no validated biomarkers are available to diagnose disease, to
monitoror predictdiseaseprogression, or to aid in assessment
of early treatment effects, despite many previous attempts to
this end [3–6].
MScan be clinicallydivided into threedifferent majorforms:
relapsing remitting (RRMS), secondary progressive (SPMS) and
primary progressive (PPMS) MS, respectively. In addition, the
clinicallyisolated syndrome (CIS) hasbeendescribedinwhich a
single episode of the symptoms of RRMS patients may be a first
indication of MS. However, it is difficult to predict which of
these individuals will experience further bouts of symptoms,
and then by definition, go on to be diagnosed with MS [7].
RRMS is characterized by a series of exacerbations that result in
varying degrees of disability from which the patient, during
the early stages at least, often recovers to a large extent. Of
the RRMS patients 60–80% will, with time, experience a gradual
decline, and become defined as SPMS. In PPMS, however, there
arenoexacerbationsasinRRMSbutratherthediseasegradually
progresses from onset [8].
Unbiased proteomics approaches have previously identi-
fied potential candidate biomarkers in different neurological
disorders [9–13]. The markers can also add biological informa-
tion that may provide new insights into disease mechanisms.
Indeed, initial unbiased proteomics studies have identified
markers which after validation have been used as diagnostic
tools in Creutzfeldt–Jakob disease [14]. Following the initial
biomarkerdiscoveryphase,definiteestablishmentofpotential
markers,requiresvalidationstudiesin independentandlarger
patientpopulations.Proteomicstechnologyislabourintensive
and at best semi-quantitative, and therefore it is necessary to
apply more automated and quantitative methods, that can
ultimately be performed in a clinical laboratory setting.
In the present work we investigated protein expression
levels in the CSF with the aim to: 1) identify markers that
differentiate between MS patients and controls and that can
potentially be used to monitor the clinical course and/or
treatment effects; 2) gain insight into pathogenesis mechan-
isms. A biomarker discovery study was initially carried out
using two-dimensional gel electrophoresis (2DE) and mass
spectrometry of CSF from 112 individuals.
From the initial unbiased proteomics discovery study, a set
of ten potential MS biomarkers was chosen for further
evaluation, selection primarily being based on their ability to
discriminate MS from control individuals, but alsoon accessto
reagents andimmunoassay feasibility. We also consideredthe
known biological functions of the proteins in order to select
markers previously known to be involved in either inflamma-
tory or degenerative processes.
Biomarker evaluation was performed using CSF samples
from an independent MS patient case–control cohort com-
prising of 209 individuals using Luminex™ immunoassay
technology [15]. In addition, 36 of the RRMS patients had
CSF sampled following treatment. These patients were thus
analyzed both before and after 6 or 12 months treatment with
the drug natalizumab (Tysabri). Natalizumab is an α-4
integrin antibody that inhibits the entrance of leucocytes
into the central nervous system, and this is an efficacious
MS treatment [16,17]. To our knowledge, ours is the first study
using CSF from MS patients in which unbiased proteomics
has been successfully followed by an independent evaluation
of selected markers. We demonstrate that the selected CSF
markers can discriminate between subjects with MS and a
control group.
2. Materials and methods
2.1.Subjects and clinical samples
All samples were collected at Karolinska University Hospital
Stockholm, Sweden, during investigation of patients with
possible multiple sclerosis, using diagnosis criteria as de-
scribed in [2]. Patient information is summarized in Table 1a,
(biomarker discovery study) Table 1b and c (evaluation study),
respectively. As controls we selected samples from persons
with other neurological diseases without having oligoclonal
IgG bands in CSF. These individuals had a variety of other
neurological signs and symptoms including, trigeminal neu-
ralgia, unspecified sensory symptoms, visual disturbance,
neurasthenia and headache. Routine blood tests, CSF analysis
and brain MRI did not reveal any sign of inflammatory disease
in this cohort of controls. In addition, in the biomarker study
a group of healthy individuals was included in the control
group, (n=9) Table 1a. The patients in the treatment study
received a monoclonal antibody natalizumab (anti-VLA-4;
Tysabri, Biogen Idec, Cambrige, MA, USA) 300 mg adminis-
trated by i.v infusion as routine treatment every fourth week.
All study enrolment followed the recommendations of the
Declaration of Helsinki and the study was approved by the
Ethics Committee of the Karolinska Institute. Oral and written
1118
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 3

information was given to the patients and confirmed consent
in writing was received before inclusion.
2.1.1.
After the lumbar puncture 10–20 ml CSF obtained from each
individual was collected in polypropylene tubes and the fluids
were centrifuged for 12 min at 350×g. The supernatants were
immediately transferred to new polypropylene tubes, placed
on dry ice and were stored at −80 °C until use as cell-free CSF.
Preparation of cell-free CSF samples
2.2.Proteomics biomarker discovery study
For a detailed description of the procedure, see our previous
publication on a related condition, Post-Polio Syndrome [18]
(biomarker discovery workflow Fig. 1). Briefly, all CSF samples
were affinity purified to remove albumin and IgG to achieve
higher sensitivity and specificity for less abundant proteins.
To adjust for different protein concentrations in the samples,
one-dimensional gel electrophoresis was followed by image
analysis and the total protein concentrations were determined.
All samples were prepared and analyzed by 2DE in a blinded
and randomized sequence and the total protein concentration
loaded was ∼600 μg/gel. Isoelectric focusing was performed
using 24 cm IPG strips pH 4-7L (GE Healthcare, Uppsala,
Sweden), and the second dimension separation was conducted
using1.0×220×200 mm12.5%TSDS-PAGE.2D-gelswerestained
with Sypro Ruby (Molecular Probes, USA) over night and finally
scanned at 100 μm resolution (Molecular Imager FX, Bio-Rad,
Hercules, CA, USA). Image files were processed using the
software PDQuest (version 7.3, Bio-Rad, Hercules, CA, USA).
Detected protein spots were then matched between gels and a
synthetic master image was prepared to represent a majority of
the protein spots present in all gelsand groupsof samples. This
was followed by image analysis and quantification (relative
integrated optical density of protein spots) using PDQuest
software and data analysis using MATLAB. Protein spots of
interest were excised from gels using an EXQUEST spot cutter
robot (Bio-Rad, Hercules, CA, USA), and transferred to 96-well
plates. Up to six protein gel spots corresponding to the same
spot identity were pooled to each well in order to facilitate the
identification of low abundant protein spots. Proteins were
digested with trypsin and the samples were further purified
using a Micro-tech Workstation and/or LC–MALDI-MS/MS
reversed-phase chromatography [18,19].
Table 1
Group/diagnosisN/groupSexNMean age Age range N (edss)/groupMedian EDSS/group EDSS range
a. Patients and diagnosis for biomarker discovery phase. N=112
HC9
OND/TH10
F
F
M
F
M
F
M
F
M
F
M
F
M
9
5
5
33.7
27.0
39.8
47.3
54.3
34.1
39.5
38.9
39.4
33.8
42.0
56.5
54.2
25–54
14–37
21–52
30–73
38–74
20–58
28–50
21–59
21–54
17–47
31–55
39–82
45–65
0
0
0
4
0
5
2
–
–
–
0.5
–
1
2
1.75
1.75
2.75
3
5
3.75
–
–
–
OND17130–2
–
0–2.5
1–3
0–6
0–2.5
1.5–4
2–4
3–6
2.5–6.5
4
CIS1410
4
RR rem 36 2828
8
6
3
8
6
3
RR rel9
SP17 1111
66
b. Patient information and diagnosis for biomarker evaluation phase, N=209
OND 58F
M
F
M
F
M
F
M
F
M
43
15
47
29
13
42.7
41.3
41.9
37.8
40.5
35.9
54.7
50.8
52.3
45.3
25–75
21–61
17–69
24–61
23–60
22–56
35–69
39–62
36–64
38–53
8
2
1.25
1
3
3
3.5
3.5
5
5.5
3.5
4.25
0–2
0–2
0–7.5
0–7
1–5.5
1.5–4.5
2.5–8
3.5–6.5
2.5–5.5
4–4.5
RR rem 76 46
28
13RR rel22
99
SP43 29
14
22
14
PP106
4
6
4
Treatment
duration
(months)
N/
group
Sex NMean age
(pre-
treatment)
Age range
(pre-
treatment)
Median
EDSS
pre
Median
EDSS
post
EDSS
range
pre
EDSS
range
post
Median EDSS
difference
post−pre
c. Patient information and diagnosis for natalizumab-treated group. N=36
69F
M
12 27F
M
5
4
43.0
35.5
39.6
41.8
30–59
26–53
22–54
28–56
4
2.5
4.5
2.5
4
2
4.5
2.75
2.5–6.5
0–5
0–7.5
1–6.5
2.5–6.5
1–5
1–8.5
0–6.5
0
0
19−0.5
08
Abbreviations: –; no data obtained, N (edss)/group; number of patients for which EDSS is available.
1119
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 4

All MS/MS data from the MALDI-TOF/TOF instrument was
acquired using the 4700 Explorer software (Applied Biosystems,
Foster City, CA, USA) allowing non-redundant and fully
automated selection of precursors for MS/MS acquisition.
Firstly, MS spectra were recorded from each sample spot
position and each spectrum was generated by accumulating
data from 1000 laser shots. Secondly, the MS spectra were
analyzed using the job-wide interpretation method. Spectral
peaks that met the threshold criteria of a signal-to-noise ratio
above 50 were selected automatically for the subsequent MS/
MS portion of the experiment. A mass filter excluding noise
peaks and matrix cluster ions was applied. Thirdly, the list
was automatically imported into the 4700 Explorer software
batch editor, and MS/MS spectra were recorded using air as the
collision gas with 1 kV collision energy setting. During MS/MS
data acquisition, 2500 shots (20 sub-spectra accumulated from
125 laser shots each) were allowed for each spectrum. External
calibration was performed using a mixture of peptides with
known masses in MS mode (4700 Proteomics Analyzer Calibra-
tion Mixture, Applied Biosystems Foster City, CA, USA). Default
calibration was used in MS/MS mode.Protein identificationwas
undertaken using database searches of MS/MS data obtained
from a MALDI-TOF/TOF instrument with Mascot as the search
enginesuppliedby Matrix Science. All searches were performed
against the human, rat, and mouse subset of standard protein
sequence databases (GenseqP, RefSeqP, PDB, PIR, SwissProt and
TREMBL). The Global Proteome Server (Applied Biosystems,
Foster City, CA, USA) was used for submitting data acquired
from the TOF/TOF for database searching (peptide mass
tolerance of 50 ppm and a fragment mass tolerance of 0.2 Da).
Oxidation on (M) and carbamidomethylation were allowed as
variable modifications.
2.3.Analysis of proteomics data
The image analysis resulted in 1499 detected and matched
spots for the whole dataset. Due to biological and experimen-
tal variations it is not possible to match every spot in all
gels. We applied two different approaches to handle missing
values, resulting in two datasets (referred to as w2 and ls,
respectively). For a more detailed description of this and other
pre-processing steps, see [18].
In order to identify a minimum number of variables with
maximal ability to discriminate between MS and controls,
Partial Least Squares (PLS) [20] modelling was applied to the
data. Two-group classification (prediction) models were built
for CIS, RR remitting, RR relapsing and SPMS samples vs the
control group, as well as a combined MS model (RR remitting,
RR relapsing and SPMS) vs the control group. Furthermore,
two additional patient groups were formed based on clinical
parameters, one low neurodegeneration (lowND) group, and
one high neurodegeneration group (high ND). The criteria for
inclusion in the lowND group were clinical diagnosis CIS and
RR, low MRI and expanded disability status scale (EDSS, see
Fig. 1 – Study design and layout. In the left panel the biomarker discovery workflow including all sampling, preparation and
analysis steps with decision-points. The arrow connects the final biomarker selection to the biomarker validation workflow
(right panel). (For more detailed information on experimental conditions see Materials and methods).
1120
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 5

[2]) below or equal to 3. This resulted in 8 patients. The
criteria for inclusion in the high ND group were clinical
diagnosis SP and EDSS score above or equal to 3.5, resulting in
11 patients. A PLS model was built to identify find variables
that separated these two groups, this model being termed the
‘ND’ model.
To optimize model parameters and select important
variables, a procedure that shares resemblance with both
cross-validation and bootstrap was devised that works as
follows: the samples are randomly divided into one training
set (2/3 of the samples) and one test set (the remaining 1/3).
A model is built on the training set, a variable importance
measure, the so-called variable importance in the projection
(VIP) is calculated for all variables and the test set is predicted
on the model. Note that the variable ranking and selection is
only performed on the training set, as inclusion of the test set
in the variable selection can result in substantial overestima-
tion of the predictive ability of the model. New models are
built on successively reduced numbers of variables as ranked
by the VIP parameter. This process is repeated for 100 random
partitions of the samples into training and test sets for models
with a varying number of PLS components (1–10). The
prediction rate is calculated as an average of the percentage
of correctly classified samples across the 100 iterations. When
using PLS, a cut-off threshold needs to be set in order to
establish which class a certain sample belongs to. Cut-off
optimization was also performed through the cross-valida-
tion/bootstrapping procedure as described below for the
minimum number of variables achieving an acceptable
prediction performance in order to obtain balanced prediction
rates between the two classes in the model [18]. The PLS
modelling wasconducted in Matlab (The Mathworks, Inc, USA)
using the PLS toolbox (Eigenvector Research, USA) (biomarker
discovery workflow Fig. 1).
For selection of the most important variables for each
comparison an overall VIP rank was formed using the whole
set of samples, and taking the maximum VIP for the two
datasets (w2 and ls). To filter out low expressed proteins,
the following filter criteria were applied to the VIP list: to be
classed as increased, the mean level in the disease group
should be above 100 ppm, with a fold change above 1, and the
spot shouldbe presentin at least 60% of patients in the disease
group. The criteria for a decrease was analogous, with the spot
present in at least 60% of control patients, the mean level
in control above 100 ppm and fold change below −1. These
filtered ranking lists were then used to guide the selection of
spots for identification with mass spectrometry (biomarker
discovery workflow Fig. 1).
2.4.Immunoassay development
Method development and sample analysis were performed
at RBM, (Austin, Texas, USA http://www.rulesbasedmedicine.
com/).
Microsphere-based multiplex assays for ten proteins
were developed using antigen-specific capture and detection
antibodies in a sandwich format (the assay for A2MG is a
competitive assay). Due to different requirements regarding
sampledilution,athree-plexassayforA1AC,A2MGandANGT,
and a seven-plex assay for contactin 1, fetuin A, fibulin 1,
NCAM 1, NrCAM, SOD1 and VDBP were developed. For ab-
breviations see legend to Fig. 5. Prior to assay CSF samples
were pre-diluted 10-fold for the three-plex and 200-fold for the
seven-plex assay. All incubations took place at room temper-
ature in the dark. In the wells of a flat-bottom microtiter plate,
a diluted mixture of capture-antibody microspheres (5 µL) was
mixedwith5 µLblockingbufferand10 µLstandard,pre-diluted
sample or control, respectively. The plate was incubated for
1 h. 10 µL biotinylated detection antibody (or for A2MG a
biotinylated antigen) was added to each well, the contents
were thoroughly mixed and the plate was incubated for 1 h.
10 µL diluted streptavidin–phycoerythrin was added to each
well, the contents were thoroughly mixed and the plate was
incubated for 30 min. A filter-membrane microtiter plate was
pre-wetted by adding 100 µL wash buffer followed by aspira-
tion using a vacuum manifold device. The reaction contents
of the plate were then transferred to the corresponding wells
of the filter plate. All wells were vacuum-aspirated and the
contents were washed twice with 100 µL wash buffer. After the
last wash, 100 µL wash buffer was added to each well and the
washed microspheres were resuspended with thorough mix-
ing.TheplatewasthenanalyzedusingaLuminex100Analyzer
(biomarker development Fig. 1).
Prior to analysis of study samples, both multiplex assays
were validated with respect to intra- and inter-run precision
(using quality control samples at three concentration levels
over the calibration range and covering the expected concen-
trations in samples), dilution linearity, spike recovery, cross-
reactivity and least detectable dose (LDD). For all proteins
inter-run precision was found to be in the range of 4.8–14.1%
(CV), except for the lowest concentration control sample of a
few proteins (A1AC, fibulin 1, NCAM 1 and VDBP) where inter-
run precision was 18.0–22.5%; CV. Spike recovery of authentic
proteins was in the range of 102–120%, except for A1AC where
a spike recovery of 73% was observed.
2.5. Analysis of immunoassay data
Prior to analysis of the data values below least detectable dose
(LDD) were set to LDD/2, where for each protein an LDD was
determined as the concentration corresponding to the mean
signal +3 standard deviations of 20 blank readings. A natural
log transformation was also applied to the data.
2.5.1.
To investigate whether any of the proteins are differentially
expressed between MS and controls, two different types of
analyses were conducted. In one analysis a linear model with
disease group and sex as fixed effects was applied individually
to each protein variable, using SAS version 9.1.3. Pairwise
comparisons for each MS subgroup vs the control group were
carried out as contrasts within the model to estimate the
relative difference in means between the two groups. In ad-
dition to a point estimate of the relative difference in means,
two-sided 95% confidence intervals were estimated and as-
sociated two-sided p-values were calculated. The level for
statistical significance was set to 0.05. No multiplicity adjust-
ments of the p-values were performed.
In the other analysis, classification (prediction) of PLS
models using all ten proteins were constructed using the pls
MS vs control
1121
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 6

package of R [21]. Two separate models were constructed, one
with RRMS patients (combining relapsing and remitting) vs
controls, and one with SPMS patients vs controls. Models were
constructed using 6 components and by setting the status of
controls to 0 and the MS groups to 1 prior to training. Predicted
values following LOO (leave-one-out) cross-validation were
classed as being either above or below a given threshold. The
performance of the models was calculated for a range of
threshold values (0–1.0, increment 0.1). Performance was
measured as Sensitivity (True Positive/(True Positive+False
Negative)) and Specificity (True Negative/False Positive+True
Negative) at that range of threshold values. The results were
plotted as a ROC curve.
2.5.2.
For the comparison of protein levels post vs pre natalizumab
treatment, patient data was modelled in SAS v9.3 using a
mixed effects model with patient as random effect and time
(post or pre) as fixed effect. This model takes into account
that observations from the same patients are correlated,
but assumes that observations from different patients are
independent. Six and 12 months duration data were mod-
elled separately. The model was applied to each protein
individually. Within the model, relative estimates of the
mean post- vs pre-treatment difference were calculated,
along with a two-sided 95% confidence interval and associ-
ated two-sided p-value. The level for statistical significance
was set to 0.05. No adjustments of the p-values for testing
many variables were performed (biomarker discovery work-
flow Fig. 1).
Post- vs pre-natalizumab treatment
3.Results
3.1.Biomarker discovery identification and selection
In the biomarker discovery phase of this study we analyzed
1499 protein spot expression profiles by employing proteomic
techniques with CSF from 76 MS individuals compared with 36
control individuals including 9 healthy controls. Table 1a
provides information regarding diagnosis, sex, age and
number of patients included in the proteomics study. Fig 1
describes the workflow in the biomarker discovery phase
(Fig.1left panel)andevaluationphase(Fig.1 rightpanel)ofthe
study. To obtain a measure of the ability of the protein spots to
separate the groups, e.g. disease vs control, PLS prediction
modelling was performed. For the combined MS (RR and SP)
model, the prediction rate was 80% with 787 proteins in MS,
and that number decreased to 72% with the top-50 spots
(Fig. 3). Models were also constructed for the MS subgroups vs
control: CIS, RR remitting phase (RR rem), RR relapsing phase
(RR rel), and SPMS, respectively. The prediction rates for CIS,
RR rem, RR rel, and SPMS at 787 proteins were 70%, 75%, 73%
and 89%, respectively. The prediction rates for the top-50 most
predictive proteins were, with the same group order, 69%, 67%,
74% and 78%, respectively (Fig. 3). A summary of the altered
protein levels from the different disease groups, CIS, RRMS
and SPMS, displayed on a 2DE master image is presented in
Fig. 2. Protein identities in the case that a positive identifica-
tion emerged from the mass spectrometry are presented in
Table 2a–d. During the biomarker discovery phase a substan-
tial number of interesting proteins were not selected for
further evaluation studies due to feasibility. The top 50 most
contributing proteins in each group were chosen for identifi-
cation by mass spectrometry (Fig. 3), however, due to low
concentration levels in the gels the success rate was ∼20%.
Table 2a–d depicts the identified proteins and the VIP ranking
associated in the different PLS models. Supplementary Table 1
provides protein information obtained from the identification
by mass spectrometry. Note that obtained protein expression
levels are based on a single isoform or post-translational
modification from a given position on the gel of a specified
protein. For some proteins, several spots shared same identity
and showed consistent expression pattern with respect to
direction and disease group (Fig. 2). Ten potential MS
biomarkers were chosen for further evaluation, based on VIP
rank, consistent expression pattern on the gels, whether
they could be identified, as well as by immunoassay feasibility
(e.g. access to antibodies) and whether there were previous
implications from the literature of their role in MS or other
neuro-inflammatory or -degenerative processes. Table 3 pre-
sents the proteins identified by name, disease group, direction
obtained in proteomic study and sequence information
from mass spectrometric analyses. Supplementary Fig. 1a–d
shows protein spot locations on the gel with the ten candidate
biomarkers underlined with a red bar.
3.2.
multiplex immunoassay
Biomarker evaluation and quantification using
To evaluate and quantify the top ten protein panel, an
independent validation patient population comprising of
151 MS patients and 58 control patients was identified, and a
different technology platform, a microbead-based multiplex
immunoassay (Luminex™), was applied to obtain more accu-
rate quantification (biomarker evaluation workflow Fig. 1).
Table 1b and c provide information regarding diagnosis, sex,
age, EDSS and number of patients included in the evaluation
study.
3.2.1.
To investigate whether any of the proteins were differentially
expressed between MS patients and controls, they were
analyzed both individually in a comparative analysis of
means, as well as in a classification model analysis, using all
ten proteins in the model build. Considering the comparisons
at the individual protein level, the mean expression levels in
CSF appear generally to be increased in both RRMS and SPMS
as compared to controls, (Table 4 and Fig. 4, Supplementary
Table2). In theRR remissiongroup(n=76) therewere smallbut
statistically significant increases of A1AC (21%), A2MG (24%)
and fibulin 1 (19%) compared to the controls (n=58). In RR
relapsing MS (n=22) there were significant increases of A1AC
(20%) and A2MG (20%), but not of fibulin 1 (p=0.092). In SPMS
(n=43) there were significant increases for A1AC (37%), A2MG
(29%) and fibulin 1 (30%). In addition, the levels of contactin
1 (27%), fetuin A (20%), VDBP (35%), and in particular of
ANGT (3 fold) were significantly higher in the CSF from SPMS
patients compared to controls. Finally, in CSF from a small
group of PPMS patients (n=10) significantly higher levels were
MS vs controls
1122
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 7

recorded for A1AC (19%) and A2MG (31%). The estimated
relative increase for ANGT in PPMS was 1.6 fold, with a 95%
confidence interval ranging from 24% decrease to 3.3 fold
increase, indicating that there might be a difference in the
PPMS group compared to the control group similar in size to
that observed for ANGT in SPMS, however this needs to be
confirmed in CSF from a larger cohort of PPMS patients. A
summary of the estimated relative differences along with 95%
confidence intervals is illustrated in Fig. 4 for all proteins and
all comparisons. The same information is tabulated together
with corresponding p-values in Table 3.
In the PLS modelling using cross-validation to estimate
prediction performance, a sensitivity of 0.72 and specificity of
0.68 was obtained with a cut-off value at 0.5 for the RRMS vs
control model (Fig. 5, left panel), and for the SPMS vs control
(Fig. 5, right panel) model the obtained sensitivity was 0.84 and
the specificity was 0.79 (cut-off value at 0.5). A set of 36 RRMS
patients independent of the model build was available and
was used as an independent test set for the RRMS model. 64%
were classified as RRMS, providing further independent value
of the potential utility of the panel. These patients were also
treated with natalizumab (see below) and analyzed in the pilot
study.
The results from both these analyses indicate that the
expression profiles in the CSF biomarker panel may provide
interesting biological information reflecting disease state.
3.3.
biomarkers
Effect of natalizumab therapy on candidate CSF
The panel of candidate biomarkers was also assessed in a pilot
study of CSF samples from natalizumab-treated patients.
Thirty-six of the RRMS patients had a CSF sample taken both
before and after treatment with natalizumab at time points 6
(9 patients) and 12 months (27 patients) (Table 1c). The mean
protein levels for the 9 patients treated for 6 months appear
generally to be decreased after treatment and A1AC, contactin
1, NCAM-1 and NrCAM reach statistical significance (Fig. 6
upper panel, Fig. 7, Supplementary Table 3) For the groups
of patients that were treated for 12 months, there were no
significant differences between post- and pre-treatment
(Fig. 6, lower panel, Fig. 7, Supplementary Table 3). It is
noteworthy that the mean CSF levels for the majority of the
protein markers (e.g. A1AC, NrCAM and contactin 1) were
higher in the 6 months group of RRMS patients than in the
12 months group before treatment, and more similar to levels
found in the overall cohort of RRMS patients. In contrast, the
pre-treatment levels of these markers in the 12 months group
were more similar to levels observed in the control cohort,
(compare control values in Supplementary Table 2 with
“before” values in Supplementary Table 3). Additionally, to
be able to estimate treatment effects without the influence
of confounding factors, CSF samples from placebo-treated
Fig. 2 – The CSF 2D master image which show results from a statistical analysis of protein levels in CSF between controls and
three subgroups of MS. Proteins that increase (triangle) or decrease (circle) are displayed in six different colors to show the
combination of subgroups (CIS and/or RR and/or SP) in which the alteration occurs. Red for the combination of CIS/RR/SP, purple
RR/SP, yellow CIS/RR, light purple CIS/SP, green RR and blue SP. Note that an increase/decrease within so-called “trains” of
spots, usually reflecting post-translational modifications of a given protein, is very consistent also regarding MS subgroup
combination. The information from this analysis and prediction models were used to guide excision of proteins for
identification by mass spectrometry.
1123
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 8

Table 2
SSPMr
(kDa)
pIVIP rankSwissProt #
(human)
Protein name
a. Identified proteins from analysis group CIS: controls vs clinically isolated syndrome
58171025.57
6806 115 5.99
18041044.52
5803102 5.51
360673 5.00
3815 107 5.08
408504.39
250167 4.73
1
4
CNTN1
A2MG
TRFE
CNTN1
CO9
CERU
A2GL
ANGT
Contactin 1
Alpha-2-macroglobulin
Serotransferrin
Contactin 1
Component C9
Ceruloplasmin
Leucine-rich alpha-2-glycoprotein
Angiotensinogen
15
17
20
26
42
57
b. Identified proteins from analysis group SP: controls vs secondary progressive MS
5114 18
3821 119
441448
5.67
5.13
5.38
1
4
6
SOD1
A2MG
PAM
Superoxide dismutase 1
Alpha-2-macroglobulin
Peptidylglycine alpha-amidating
monooxygenase isoform b
Agrin
Neuronal cell adhesion molecule
Kallikrein 6
Neuronal cell adhesion molecule
Nidogen-2 precursor
Alpha-1-antichymotrypsin
Amyloid beta A4 protein (fragment)
Pigment epithelium-derived factor
Apolipoprotein E
Chitinase-3-like protein 1
Contactin 1
4108
3727
7112
4719
2218
513
1109
4418
5223
6307
5803
21 5.28
5.16
6.34
5.37
4.91
4.37
4.55
5.43
5.70
5.90
5.51
7 AGRN
NRCAM
KLK6
NRCAM
NID2
AACT
A4
PEDF
APOE
CHI3L1
CNTN1
101
19
100
31
69
23
50
33
42
102
11
13
14
18
34
37
38
41
44
48
c. Identified proteins from analysis groups: controls vs relapsing/remitting MS bold text equals relapsing and non-bold equals remitting group
(RR rel and RR rem)
4304355.261
2501 67 4.732
58121015.634
68061155.995
620629 5.867
2721924.878
6307425.909
4315415.47 13
2215294.90 14
1516674.6914
331443 5.11 18
431638 5.4819
3821 1195.13 21
4307 435.2922
3722975.11 27
1832103 4.5834
230243 4.72 35
340849 5.09 37
4705995.22 37
APOE
ANGT
CNTN1
A2MG
SODE
FBLN1
CHI3L1
HAPTO
AMBP
CNDP1
HEMO
FETUA
A2MG
APOE
CERU
NCA11
CO3
GELS
NRCAM
Apolipoprotein E
Angiotensinogen
Contactin 1
Alpha-2-macroglobulin
Extracellular superoxide dismutase
Fibulin-1
Chitinase-3-like protein 1
Haptoglobin
Alpha-1-microglobulin
Carnosinase 1
Hemopexin
Fetuin A
Alpha-2-macroglobulin
Apolipoprotein E
Ceruloplasmin
Neural cell adhesion molecule 1
Complement C3 fragment
Gelsolin
Neuronal cell adhesion molecule
d. Identified proteins from analysis group “ND”: low level of neurodegenerative MS vs high level of neurodegenerative MS, defined by clinical criteria
(see “Materials and methods”)
2501 67 4.731
1804 1044.522
251559 4.74 6
1504674.4715
561886 5.7321
48041065.22 32
4211325.29 33
4514655.40 35
140648 4.5144
ANGT
NCA11
A1AT
AACT
GELS
CERU
APOE
APOE
CO4A
Angiotensinogen
Neural cell adhesion molecule 1
Alpha-1-antitrypsin
Alpha-1-antichymotrypsin
Gelsolin
Ceruloplasmin
Apolipoprotein E
Apolipoprotein E
Complement C4-A
1124
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 9

patientswouldbe neededin thestudy. Withthe currentdata it
is not possible to separate events that change with time from
the effects of the treatment, and it is far too early to make any
conclusions about natalizumab effects on the biomarker
panel.
4.Discussion
We have identified and evaluated a panel of 10 proteins in CSF
as candidate biomarkers for MS for diagnosis and monitoring
of disease stage/progression in MS.
The study aims were to: 1) identify and evaluate markers
that can differentiate between MS and control subjects and
that can potentiallybe usedto monitorthe clinicalcourseand/
or treatment effects: 2) potentially gain further insight into
pathogenesis mechanisms. From the biomarker discovery
phase using proteomic tools we identified proteins using
an unbiased design from which a selection of 10 interesting
proteins was made to build an immunoassay for thorough
quantification and further evaluation (Figs. 1–3 and Table 3).
The results from the proteomic discovery part of this study
suggest that there is a pronounced heterogeneity within the
MS patient group, as indicated by the PLS classification model
(Fig. 3, Table 2a–d) results. This is in agreement with the
hypothesis that MS is a multifactorial disease with genetic
and environmental influence and a complex biology. Indeed,
this is manifested by the large variation in degree of severity
and disease progression rate in individual patients during
the disease course. A number of interesting proteins from the
biomarker discovery phase of our study were not included into
the immunoassay (Table 2a–d), and further studies of these
proteins are warranted to understand the implications of the
findings. However, the scopes for the current studies were to
attempt to establish biomarkers useful in the clinical practice,
e.g. clinical studies and trials.
4.1. Immunoassay analysis of protein expression
The results obtained from the quantitative immunoassay
revealed clear differences between disease stages of MS. RRMS
patients exhibited modest but statistically significant
increases of A1AC, A2MG and fibulin 1 in their CSF compared
to the controls. There were only minor differences between
RRMS patients in remission and those with ongoing relapse at
time of CSF sampling (Fig. 4 and Supplementary Table 2).
However, the most prominent differences to controls in
the candidate biomarker levels were demonstrated in the
CSF from SPMS patients for which a pattern emerged with
increased levels in 7 of the 10 proteins in the biomarker panel.
Fig. 3 – Results from the model optimization procedure, which starts with all variables (spots) included and successively drops
the ones that are least important to the model. Prediction rate on Y-axis and number of proteins on X-axis, increasing protein
values from left to right. The prediction rate is the percentage correctly predicted samples from the test set, calculated as an
average of the 100 random partitions. This procedure was applied to each of two datasets resulting from two methodsto handle
missing values (w2 and ls), the prediction curve showed in plot is for the dataset giving the best result for each model.
1125
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 10

Of the 3 markers in common with RRMS (A1AC, A2MG,
fibulin 1), the most prominent in SPMS was ANGT. In addition,
CSF levels of contactin 1, fetuin A, VDBP and A1AC were higher
in SPMS compared to controls and RRMS patients. PPMS
patients had only increased CSF levels of 2 of the proteins in
the biomarker panel (A1AC and A2MG), both in common with
SPMS and RRMS. However, note that no PPMS samples were
includedin the biomarker discovery phase of this study. A2MG
was present at similar levels in all MS subgroups (20–31%
increasevs controls). Of particularnote is the significant 3-fold
increase of ANGT CSF levels in SPMS. As indicated by the 95%
confidence interval, there may be a similar increase in the
PPMS group, however the confidence interval includes 1, and
thus it cannot be excluded that there is no difference or a
small decrease. In addition, ANGT was the highest ranked
protein from analysis of the ND group (Table 2d), implying a
role during neurodegeneration. ANGT is involved in the renin/
angiotensin system (RAS) that regulates blood pressure in the
body. Interestingly, Platten et al reported recently that the RAS
is decreased in brain lesions of MS and seems to play a very
important role in driving autoimmunity in both MS and in EAE
models [22]. In the MOG-EAE animal model inhibition of the
activating enzymes angiotensinogen converting enzyme
(ACE) and renin improved the disease course in the animals
[23]. Other studies have also demonstrated elevated activity in
ACE in serum samples from MS individuals [24]. ANGT may be
important for the maintenance and selectivity of the blood
brain barrier (BBB), as it is being produced by astrocytes and
Table 3 – Protein selected for immunoassay. With SSP number from the 2DE gel, SwissProt accession number, protein name,
disease group, direction in disease vs control and sequence from MSMS.
SSPSwissProt
accession
(human)
Protein name Disease
group
Direction in disease
vs controls
Sequence
5114
2721
SOD1
FBLN1
Superoxide dismutase 1
Fibulin-1
SP
RR
Decrease
Decrease
TLVVHEKADDLGK (SEQ ID NO.1)
CLAFECPENYR (SEQ ID NO.2)
TGYYFDGISR (SEQ ID NO.3)
IIEVEEEQEDPYLNDR (SEQ ID NO.4)
AIGYLNTGYQR (SEQ ID NO.5)3821
6806
3727
4705
4719
A2MG Alpha-2 macroglobulinSP, CIS, RRDecrease
NRCAM Neuronal cell adhesion
molecule precursor
SP, (RR) DecreaseDYIIDPR (SEQ ID NO.6)
ENIVIQCEAK (SEQ ID NO.7)
AETYEGVYQCTAR (SEQ ID NO.8)
GAAVSNNIVVRPSR (SEQ ID NO.9)
TVYKNFEK (SEQ ID NO.10)
TLQIIHVSEADSGNYQCIAK (SEQ ID NO.11)
NALGAIHHTISVR (SEQ ID NO.12)
KIDGDTIIFSNVQER (SEQ ID NO.13)
IDGDTIIFSNVQER (SEQ ID NO.14)
QPEYAVVQR (SEQ ID NO.15)
LSPYVNYSFR (SEQ ID NO.16)
SLPSEASEQYLTK (SEQ ID NO.17)
YIVSGTPTFVPYLIK (SEQ ID NO.18)
NAPTPQEFR (SEQ ID NO.19)
FIVLSNNYLQIR (SEQ ID NO.20)
DIQVIVNVPPTIQAR (SEQ ID NO.21)
HPNSPLDEENLTQENQDR (SEQ ID NO.22)
EQLSLLDR (SEQ ID NO.23)
EQLSLLDRFTEDAK (SEQ ID NO.24)
RLYGSEAFATDFQDSAAAK (SEQ ID NO.25)
LYGSEAFATDFQDSAAAK (SEQ ID NO.26)
LYGSEAFATDFQDSAAAKK (SEQ ID NO.27)
DLDSQTMMVLVNYIFFK (SEQ ID NO.28)
WEMPFDPQDTHQSR (SEQ ID NO.29)
MEEVEAMLLPETLKR (SEQ ID NO.30)
WRDSLEFR (SEQ ID NO.31)
EIGELYLPK (SEQ ID NO.32)
ADLSGITGAR (SEQ ID NO.33)
NLAVSQVVHK (SEQ ID NO.34)
AVLDVFEEGTEASAATAVK (SEQ ID NO.35)
ITLLSALVETR (SEQ ID NO.36)
EHAVEGDCDFQLLK (SEQ ID NO.37)
FIPLIPIPER (SEQ ID NO.38)
1832
1804
NCA11 Neural cell adhesion molecule 1ND, RRDecrease
513
1504
AACTAlpha-1-antichymotrypsin SP, ND Increase
4316
5803
5812
2501
FETUA
CNTN1
Fetuin A
Contactin 1
RR
CIS, RR, (SP)
Increase
Increase
ANGTAngiotensinogen ND, RRDecreaseAAMVGMLANFLGFR (SEQ ID NO.39)
VLSALQAVQGLLVAQGR (SEQ ID NO.40)
QPFVQGLALYTPVVLP (SEQ ID NO 41)
KFPSGTFEQVSQLVK (SEQ ID NO.42) 3516VDBPVitamin D binding proteinSP, RRIncrease
1126
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 11

influences endothelial cells within the blood brain barrier [25].
One explanation for the increased levels of ANGT observed in
CSF from SPMS patients may thus be that there is increased
expression of ANGT when the brain is trying to maintain
integrity of, or trying to repair, an injured BBB. Moreover, an
interesting observation is that permanent activation of the
RAS impairs cognitive functions in transgenic mice [26].
Fibulin 1 is an extracellular matrix protein that has been
detectedinCSF[27]andwhichcanbindandmodulateAmyloid
Precursor Protein (APP) [28]. The involvement of APP in
Alzheimer's disease (AD) is known and an interesting obser-
vation from our studyis that a fragmentof about 24 kDa of APP
is decreased in MS patients (Table 2b). Indeed, in a 10 year
follow-up study in MS it was reported that BACE1 activity was
reduced over time, resulting in decreased levels of APP [29].
A2MG and A1AC were statistically significantly increased
to somewhat similar degree in all forms of MS compared to
controls. A2MG has been detected in amyloid plaques in
Alzheimer's disease, and interestingly a possible contribution
of A2MG in neurodegenerative disorders was recently pub-
lished based on observations in a mouse model [30]. A2MG is a
general inhibitor of metalloproteinases, and levels of free
A2MG in serum have been used as a surrogate marker of
metalloproteinase activity [31–33]. As metalloproteinases are
significant mediators of inflammation, including EAE and MS
[34–36], this biomarker is of particular interest in that respect.
A2MG is involved in acute phase responses and signaling,
which is also the case for A1AC.
A1AC has been shown to be involved in the development of
Alzheimer's disease (AD)-like pathology in a mouse model [37]
and is also overexpressed in AD brain and affected by the
APOE4 genotype [38]. Indeed, expression has been detected in
astrocytes located in the plaques and the adjacent surround-
ing tissue [39]. Moreover, previous studies have shown that
A1AC is overexpressed also in other, non-AD, tauopathy
disorders and can induce tau phosphorylation and apoptosis
in neurons [40]. Several studies have demonstrated elevated
levels of A1AC in CSF of AD patients [41,42] and one group
Table 4 – Results from comparisons of MS subgroups RR rem, RR rel, SP and PP to controls including protein, comparison,
ratio, CI and p-value. p-value in bold <0.05.
VariableComparisonRatio Lower 95% CIUpper 95% CIp-value
A1AC RR rem vs controls
RR rel vs controls
SP vs controls
PP vs controls
RR rem vs controls
RR rel vs controls
SP vs controls
PP vs controls
RR rem vs controls
RR rel vs controls
SP vs controls
PP vs controls
RR rem vs controls
RR rel vs controls
SP vs controls
PP vs controls
RR rem vs controls
RR rel vs controls
SP vs controls
PP vs controls
RR rem vs controls
RR rel vs controls
SP vs controls
PP vs controls
RR rem vs controls
RR rel vs controls
SP vs controls
PP vs controls
RR rem vs controls
RR rel vs controls
SP vs controls
PP vs controls
RR rem vs controls
RR rel vs controls
SP vs controls
PP vs controls
RR rem vs controls
RR rel vs controls
SP vs controls
PP vs controls
1.21
1.20
1.37
1.19
1.24
1.20
1.29
1.31
0.97
0.85
2.96
1.59
1.14
1.31
1.27
1.10
1.14
1.15
1.20
1.06
1.19
1.19
1.30
1.03
1.09
1.00
1.13
1.12
1.11
1.12
1.08
1.14
1.11
1.08
1.14
0.89
1.09
0.98
1.36
1.01
1.11
1.07
1.25
1.02
1.12
1.03
1.13
1.06
0.66
0.50
1.92
0.76
0.93
0.98
1.01
0.74
0.99
0.94
1.02
0.80
1.03
0.97
1.11
0.78
0.90
0.76
0.91
0.77
0.88
0.79
0.82
0.71
0.98
0.90
0.99
0.70
0.91
0.76
1.11
0.72
1.31
1.35
1.51
1.40
1.39
1.40
1.46
1.62
1.41
1.46
4.56
3.33
1.39
1.75
1.60
1.63
1.32
1.42
1.42
1.41
1.36
1.45
1.52
1.35
1.32
1.32
1.41
1.63
1.42
1.58
1.42
1.83
1.26
1.29
1.32
1.13
1.30
1.27
1.67
1.43
<.0001
0.0023
<.0001
0.0299
0.0001
0.0228
0.0001
0.0129
0.8696
0.5600
<.0001
0.2148
0.1982
0.0663
0.0443
0.6251
0.0715
0.1825
0.0323
0.6859
0.0163
0.0915
0.0014
0.8541
0.3741
0.9877
0.2799
0.5621
0.3770
0.5268
0.5904
0.5814
0.0943
0.4211
0.0630
0.3449
0.3369
0.9053
0.0029
0.9375
A2MG
ANGT
Contactin 1
Fetuin A
Fibulin 1
NCAM 1
NrCAM
SOD1
VDBP
1127
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 12

reported very interestingly that the levels of A1AC in both
plasma and CSF correlated with degree of disease severity in
AD patients [42].
The levels of VDBP were increased in SPMS. Vitamin D has
previously been in focus in MS research, and differences in
levels in MS patients have been reported, whereby levels of 25-
Fig. 4 – Results from the immunoassay data analysis of each protein individually, comparing MS subgroups vs the control
group. The plot shows estimates of the relative differences (ratios) of protein concentrations for RR rem RR rel, SPMS and
PPMS vs the control. Whiskers indicate upper and lower 95% confidence limits. Note the different scale for angiotensinogen
with Y-axis values between 0.5 and 5.0. Protein abbreviations are: A1AC (alpha-1-antichymotrypsin),
A2MG (alpha-2-macroglobulin), ANGT (angiotensinogen), NCAM1 (neural cell adhesion molecule 1), NrCAM (neuronal cell
adhesion molecule), SOD1 (superoxide dismutase 1), and VDBP (Vitamin D binding protein).
Fig. 5– ReceiverOperatorCharacteristic(ROC)curves obtained fromtwo PLS models.ROCcurvesplot sensitivity vs specificityfor
a given predictive model, where the performance is analyzed by adjusting some parameter, in this case classification cut-off
threshold, such that individual samples predicted (from all other samples) to be above the cut-off threshold are classed as MS,
whereas samples predicted below are classed as controls. Left panel: controls vs RRMS model; Right panel: controls vs SPMS
model. The solid diagonal line indicates the portion of the graph, above which a test point is informative, that is to say confers
more information about the sample than flipping a coin.
1128
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 13

hydroxyvitamin D(3) and the active form 1,25-dihydroxyvita-
min D(3) were decreased during RRMS and especially during
relapse [43,44]. The increased levels of VDBP may thus explain
the decreased levels of Vitamin D in MS. In addition, we have
observed increased levels of phosphorylated VDBP in CSF from
RRMS and SPMS patients (http://www.wipo.int/pctdb/en/wo.
jsp?wo=2007008158) further supporting the hypothesis that
VDBP may have an important role in the etiology of MS.
Fetuin A is a protein involved in regulating calcium levels
and is decreased during inflammation [45,46], and the levels
are decreased in patients with rheumatoid arthritis (RA) [47]. It
has also been reported that expression levels were decreased
in CSF of individuals with CIS [48]. However, in SPMS we
observed a significant increase between 2 and 42% (Fig. 4),
as indicated by the 95% confidence interval (Table 3), suggest-
ing that the function may be separated in SPMS from the
inflammation events occurring in RA and other systemic
diseases including CIS.
Although the proteomics data from the discovery phase of
our study suggested differently expressed levels of SOD-1 in
CSF from MS patients, the immunoassay data showed no
significant difference in CSF levels of SOD-1 in any of the MS
groups compared to controls. SOD-1 normally acts as a free
radical scavenger but in familial Amyotrophic lateral sclerosis
(ALS),thereisamutationthatresultsinatoxicgainoffunction
explainingtheproteinaccumulationandcelldeaththatoccurs
in about 20% of the cases [49]. Interestingly, a decreased
activity of SOD1 has been shown in AD, PD, ALS and
Huntington's disease [50]. Moreover, oxidative damage and
accumulation of SOD1 may contribute to the sporadic forms of
ADandPD[51].TheCSFlevelsofcontactin1wereclosetobeing
significantly increased in relapsing RRMS (∼30%) and SPMS
(∼25%) but with large variance within the group. Contactin 1
and A2MG may also qualify as candidate biomarkers for early
diagnosis of MS, indicated by high VIP ranking in the CIS
group (Table 2a) but this needs to be further evaluated using
a new CIS patient cohort. Moreover, contactin 1 is abundantly
expressed in demyelinated axons in MS patients [52].
The biological significance of observed changes of biomar-
kersduringtreatment of RRMSpatientswithnatalizumab is so
far not fully understood. The treatment study did not include
placebo control individuals and the interpretation is conse-
quently difficult since treatment effects cannot be separated
from time effects in a post- vs pre-treatment comparison.
Moreover, the baseline mean levels for the patient group
treated for 6 months were more similar to the RRMS mean
levels, and the baseline mean levels for the patient group
treated for 12 months were generally lower, being more
similar in range to the control group, which also complicates
the interpretation.
Taken together, there is significant evidence in the
literature to associate the biomarkers that we have discovered
Fig. 6 – Results from the immunoassay data analysis of natalizumab-treated patients. The plot shows estimates of the relative
differences (ratios) of protein concentrations for post- vs pre-treatment. Whiskers indicate upper and lower 95% confidence
limits. The top panel shows results from the analysis of the patients treated for 6 months and the lower panel shows results
from the patients treated for 12 months (abbreviations see legend to Fig. 5).
1129
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 14

1130
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2

Page 15

to plausible disease-related mechanisms. It is of particular
note that the biomarkers associated with RRMS and SPMS
patients differed somewhat, supporting that the pathophys-
iology of these two disease stages are immunologically
distinct, with inflammation being more prominent in RRMS
and nerve degeneration being more prominent in SPMS. We
demonstrate that there is differential expression of the
biomarkers between RRMS and SPMS and controls, respec-
tively, using two types of analyses.
Itmaythereforebe usefultouse thispanelofbiomarkersas
a complement to the diagnostic instruments existing today
and for monitoring patients during treatment. Moreover,
these markers may also be tested in different primarily
neurodegenerative disorders such as PD, AD and ALS. The
results from the evaluation study using a multiplex immuno-
assay in a new cohort of patients and control subjects suggest
that several of the selected candidate biomarkers are differ-
entially expressed in the CSF of MS patients. The observation
of changes in the protein expression profile pre- vs post-
treatment with natalizumab also supports that proteomics
approaches can lead to new biological targets and increased
our understanding of biochemical processes occurring in MS
(Figs.6,7). Wedonotexcludethat inclusionofotherbiomarker
candidates (e.g. agrin, ApoE, A1AT, nidogen 2, see Table 2a–d)
may improve the usefulness of a MS biomarker panel,
although specific and reliable antibodies need to be developed
[53].
However, to be able to establish the usefulness of the panel
to monitor treatment effects and disease progression, com-
plementary studies with control individuals, ideally placebo-
treated, would be required.
Application of a panel of biomarkers opens the opportunity
to monitor different biological processes. In a complex disease
like MS, considered as an autoimmune disease, in which an
ongoing neuroinflammation and neurodegeneration are pres-
ent it is likely that by studying several proteins, a better
description of the disease state is obtained rather than with a
single protein. By a targeted approach it may be possible to
identifyanddevelop differentbiomarkerpanelsthatrepresent
the disease stages that could be used to monitor the disease
progression. It is also worth noting that the obtained values
from proteomics and immunoassay differed, this is most
likely dependent on that proteomic analysis can measure one
single isoform of the protein whereas antibodies measure all
protein forms.
In conclusion, to our knowledge this is one of the largest
2DE based proteomics studies of CSF from MS patients
followed by a validation study using independent patients as
well as an independent technology that has been performed to
date. By employing an unbiased proteomics approach we have
successfully identified proteins that may be involved in MS
disease maintenance and progression. This unique set of
biomarkers may potentially be used to improve the clinical
diagnosis and to monitor disease progression or treatment
effects in clinical trials of neurodegenerative disorders. The
classification performance we observe is approximately the
same as existing components of the MS diagnostic triage, and
this test might have additional predictive value when used in
conjunction with existing diagnostic tests, as well as potential
utility in monitoring and selecting treatment options. More-
over, to further develop and to increase the sensitivity of
the MS biomarker-signature in CSF identified in this study,
further addition of new proteins to the panel may increase the
specificity.
Acknowledgements
We would like to express our gratitude to all the patients that
have been involved in these studies.
Conflict of interest: There are no other known conflicts of
interest besides a patent application (reference number
103145-1 US) entitled “Diagnostic Method”. This application
was filed in US 09062009.
Appendix A. Supplementary data
Supplementary data associated with this article can be found,
in the online version, at doi:10.1016/j.jprot.2010.01.004.
R E F E R E N C E S
[1] Garcia Merino A, Blasco MR. Confirming the MS diagnosis. Int
MS J 2007;14(2):58–63.
[2] McDonald WI, et al. Recommended diagnostic criteria for
multiple sclerosis: guidelines from the International Panel on
the diagnosis of multiple sclerosis. Ann Neurol 2001;50(1):
121–7.
[3] Vladic A, et al. Cerebrospinal fluid and serum protein levels of
tumour necrosis factor-alpha (TNF-alpha) interleukin-6 (IL-6)
and soluble interleukin-6 receptor (sIL-6R gp80) in multiple
sclerosis patients. Cytokine 2002;20(2):86–9.
[4] Adam P, et al. CSF and serum orosomucoid (alpha-1-acid
glycoprotein) in patients with multiple sclerosis: a
comparison among particular subgroups of MS patients. Clin
Chim Acta 2003;334(1–2):107–10.
[5] Teunissen CE, Dijkstra C, Polman C. Biological markers in CSF
and blood for axonal degeneration in multiple sclerosis.
Lancet Neurol 2005;4(1):32–41.
[6] Norgren N, et al. Neurofilament and glial fibrillary acidic
protein in multiple sclerosis. Neurology 2004;63(9):1586–90.
[7] Miller D, et al. Clinically isolated syndromes suggestive of
multiple sclerosis, part I: natural history, pathogenesis,
diagnosis, and prognosis. Lancet Neurol 2005;4(5):281–8.
[8] Leary SM, Porter B, Thompson AJ. Multiple sclerosis: diagnosis
and the management of acute relapses. Postgrad Med J
2005;81(955):302–8.
[9] Ray S, et al. Classification and prediction of clinical
Alzheimer's diagnosis based on plasma signaling proteins.
Nat Med 2007;13(11):1359–62.
[10] Schulze-Topphoff U, et al. Activation of kinin receptor B1
limits encephalitogenic T lymphocyte recruitment to the
central nervous system. Nat Med 2009;15(7):788–93.
Fig. 7 – Individual changes of protein concentrations before and after treatment with natalizumab for 6 months (left hand side of
each graph) and 12 months (right hand side of each graph) (abbreviations see legend to Fig. 5).
1131
J O U R N A L O F P R O T E O M I C S 7 3 ( 2 0 1 0 ) 1 1 1 7 – 1 1 3 2