A Subset of Cerebrospinal Fluid Proteins from a Multi-Analyte Panel Associated with Brain Atrophy, Disease Classification and Prediction in Alzheimer’s Disease
A Subset of Cerebrospinal Fluid Proteins from a Multi-Analyte Panel Associated with Brain Atrophy, Disease Classification and Prediction in Alzheimer's Disease
Wasim Khan 0 1
Carlos Aguilar 0 1
Steven J. Kiddle 0 1
Orla Doyle 0 1
Madhav Thambisetty 0 1
Sebastian Muehlboeck 0 1
Martina Sattlecker 0 1
Stephen Newhouse 0 1
Simon Lovestone 0 1
Richard Dobson 0 1
Vincent Giampietro 0 1
Eric Westman 0 1
Andrew Simmons 0 1
Alzheimer's Disease Neuroimaging Initiative 0 1
0 1 King's College London, Institute of Psychiatry, Psychology & Neuroscience , London , United Kingdom , 2 NIHR Biomedical Research Centre for Mental Health , London , United Kingdom , 3 NIHR Biomedical Research Unit for Dementia , London , United Kingdom , 4 Department of Neurobiology, Care Sciences and Society, Karolinska Institutet, Stockholm, Sweden, 5 Laboratory of Behavioural Neuroscience, National Institute on Aging, National Institutes of Health , Baltimore , Maryland, United States of America, 6 Department of Psychiatry, University of Oxford , Oxford , United Kingdom
1 Editor: Stephen D Ginsberg, Nathan Kline Institute and New York University School of Medicine, UNITED STATES
In this exploratory neuroimaging-proteomic study, we aimed to identify CSF proteins associated with AD and test their prognostic ability for disease classification and MCI to AD conversion prediction. Our study sample consisted of 295 subjects with CSF multi-analyte panel data and MRI at baseline downloaded from ADNI. Firstly, we tested the statistical effects of CSF proteins (n = 83) to measures of brain atrophy, CSF biomarkers, ApoE genotype and cognitive decline. We found that several proteins (primarily CgA and FABP) were related to either brain atrophy or CSF biomarkers. In relation to ApoE genotype, a unique biochemical profile characterised by low CSF levels of Apo E was evident in ε4 carriers compared to ε3 carriers. In an exploratory analysis, 3/83 proteins (SGOT, MCP-1, IL6r) were also found to be mildly associated with cognitive decline in MCI subjects over a 4-year period. Future studies are warranted to establish the validity of these proteins as prognostic factors for cognitive decline. For disease classification, a subset of proteins (n = 24) combined with MRI measurements and CSF biomarkers achieved an accuracy of 95.1% (Sensitivity 87.7%; Specificity 94.3%; AUC 0.95) and accurately detected 94.1% of MCI subjects progressing to AD at 12 months. The subset of proteins included FABP, CgA, MMP-2, and PPP as strong predictors in the model. Our findings suggest that the marker of panel of proteins identified here may be important candidates for improving the earlier detection of AD. Further targeted proteomic and longitudinal studies would be required to validate these findings with more generalisability.
Funding: Data collection and sharing for this project
were funded by the ADNI (grant U01 393 AG024904
from the National Institutes of Health). This study was
supported by funds from the NIHR Biomedical
Research Centre for Mental Health and NIHR
Biomedical Research Unit for Dementia at the South
London and Maudsley NHS Foundation Trust and
Institute of Psychiatry, Kings College London, and
Alzheimer’s Research UK. This research was also
supported by grants P30 AG010129 and K01
AG030514 from the National Institutes of Health and
by the Dana Foundation.
Competing Interests: The authors have declared
that no competing interests exist.
Alzheimer’s disease (AD) is a progressive neurodegenerative disorder pathologically
characterised by lesions of misfolded proteins, the loss of synapses and an overall reduction in brain
volume. There is accumulating evidence to suggest that the clinical symptoms of the disease are
preceded by a long presymptomatic phase (~15–20 years) of abnormal β-amyloid (Aβ)
aggregation in the form of extracellular senile plaques [1,2]. The neuropathology of the disease is
associated with the development of neurofibrillary tangles prior to the onset of cognitive
impairment and the subsequent emergence of full-blown dementia [3,4]. The failure of several
clinical trials assessing therapeutic strategies to target amyloid deposition has led to the
impetus to discover biomarkers earlier in the AD pathological cascade prior to the development of
One method is to study structural neuroimaging biomarkers of AD which have been
advocated for use in early diagnosis , as well as for predicting disease progression in a prodromal
form of the disease known as Mild Cognitive Impairment (MCI) . Another rich source of
biomarkers can be found in analytes from cerebrospinal fluid (CSF), particularly,
concentrations of Aβ142, p-tau181 and t-tau which reflect biochemical changes associated with Aβ
deposition, neurofibrillary tangle formation, and neuronal cell death [7,8].
Several neuroimaging studies have since found that the combined use of MRI measures
from regions affected in AD and CSF biomarkers can provide mutually complimentary
information for disease classification and prediction [9,10]. Nevertheless, there still remains a
substantial overlap in CSF biomarker concentrations between AD and cognitively normal (CN)
individuals with an increased risk of developing the disease . Moreover, additional
biomarkers are still required to understand the exact temporospatial relationship between Aβ
deposition and tau neurodegeneration during different stages of the disease pathophysiology.
Early genetic and in-vivo experimental studies have suggested that markers of inflammation,
microglial activity and synaptic function may be important for reflecting biochemical changes
associated with the Aβ toxicity and tau neurodegeneration [12,13]. While some proteomic
studies using multiplex platforms have identified a number of protein candidates detected in
AD [14–16], few have been validated and tested in relation to well-established neuroimaging
endophenotypes of AD pathology. Discovering proteins in relation to established measures of
disease pathology may yield biologically important peripheral signatures associated with
mechanisms early in the disease.
In this study we aimed to discover CSF proteins associated with AD pathophysiology by
testing the multiplex panel with established neuroimaging measures, CSF biomarkers of AD,
Apolipoprotein E (ApoE) genotype and cognitive decline. Most importantly, we aimed to
identify a subset of proteins from the multiplex panel in order to test its diagnostic utility with
existing AD biomarkers for disease classification and MCI to AD conversion prediction at
Data used in this study was obtained from the ADNI database (adni.loni.ucla.edu). ADNI was
launched by the National Institute of Ageing (NIA) and is a multicenter project supported by
private pharmaceutical companies, and non-profit organisations for the development of
biomarkers in monitoring disease progression in MCI and AD . ADNI subjects aged 55–90
from over 50 sites across the U.S and Canada participated in the research (for further
information, see www.adni-info.org). Written informed consent was given from all participants in the
study and prior ethics committee approval was obtained from each participating site. A total of
295 subjects with baseline data that included structural imaging and multiplex CSF samples
were available for analysis and consisted of 142 subjects with MCI, 65 patients with AD, and 88
healthy control subjects.
CSF protein measurements
CSF Aβ1–42, T-tau and P-tau were measured at the ADNI Biomarker Core laboratory at the
University of Pennsylvania Medical Center, using the multiplex xMAP Luminex platform
(Luminex, Austin, TX, USA) with the INNOBIA AlzBio3 kit (Innogenetics, Ghent, Belgium)
CSF multiplex proteomic samples were measured for levels of 159 analytes using the
Human Discovery Multi-Analyte Profile (MAP) 1.0 panel and Luminex 100 platform
developed by Rules Based Medicine, Inc. (RBM), (Austin, TX) . This panel is based upon
multiplex immunoassay technology to measure a range of inflammatory, metabolic, lipid, and other
disease relevant proteins. The protocol used to quantify CSF analytes is described in detail
elsewhere [21,22]. Of the 159 analytes, only those with <10% of missing values were quantifiable
leaving 83 in total for analysis. The remaining 76 analytes were mostly below the assay
detection limit, or had other assay limitations. Each analyte has an individual standard curve with
between 6–8 reference standards. Each plate is run with 3 levels of QCs (low, medium and
high) for each analyte. A total of 16 of the CSF samples were retested using a separate never
before thawed replicate aliquot on the fifth of the five 96 well plates to provide blinded
test/retest quality control data. Assays are qualified based on least detectable dose, precision,
crossreactivity, dilutional linearity, spike recovery (assessment of accuracy), and test/re-test
performance. Cross validation to alternative methods is reported for some assays where feasible.
Further information on the process, aliquoting and storage of analytes is described in the ADNI
Biomarker Core Laboratory Standard Operating Procedures
Further assay documentation and validation reports are available from Myriad RBM (www.
myriadrbm.com). Distributions of data for individual CSF proteins were checked for normality
using Box-Cox methods and, when appropriate, transformed to approximate a normal
distribution. Information regarding the biological preparation of CSF samples and quality control
criteria of the RBM Human Discovery MAP panel can be found on the ADNI websites [23,24].
A complete list of the analytes is given in S1 Table.
Magnetic Resonance Imaging Data Acquisition and Analysis
Structural MRI images (at 1.5T) were acquired at multiple ADNI sites across the US and
Canada based on a standardized protocol . The imaging protocol included a high resolution
sagittal 3D T1-weighted MPRAGE volume (voxel size 1.1 × 1.1 × 1.2 mm³). The MPRAGE
volume was acquired using a custom pulse sequence specifically designed for the ADNI study to
ensure compatibility across scanners . Full brain and skull coverage was required for all
MR images according to previously published quality control criteria [27,28].
Image analysis was carried out using the Freesurfer image analysis pipeline (version 5.1.0)
to produce 34 regional cortical thickness and 23 subcortical volumetric measures as previously
described [29,30]. All volumetric measures from each subject were normalized by the subject’s
intracranial volume while cortical thickness measures were used in their raw form .
Measures of hippocampal and entorhinal cortex volume were selected as key a priori regions to
reflect AD pathology. A previously validated MRI-based marker of AD and MCI known as
SPARE-AD (Spatial Pattern of Abnormalities for Recognition of Early AD) was also used as a
neuroimaging marker of AD. Individualised scores of diagnostic and predictive value were
used for analysis. A complete list of regional MRI measures is given in S2 Table. Details of this
particular method have been widely published elsewhere [9,32,33].
Firstly, the RBM panel of CSF proteins were tested in relation to regional MRI measurements
(hippocampal and entorhinal volume), an MRI-based measure known as SPARE-AD score
and CSF biomarkers (Aβ142, P-tau181, T-tau) using a Spearman rank partial correlation test.
This test was adjusted for covariates including age, gender, years of education and ApoE E4
genotype. Secondly, CSF proteins from the RBM panel were also tested in relation to different
ApoE polymorphisms (ε2 carriers, ε3 carriers and ε4 carriers) using a generalized linear model
adjusting for age, gender and years of education. Thirdly, to test the effect of CSF proteins on
longitudinal MMSE score, we used a linear mixed model approach. Global MMSE score was
used as the response variable and the time from baseline visit in months, CSF protein from the
RBM panel, age, sex, years of education and ApoE ε4 genotype were included as fixed effects.
Models contained a random intercept and slope. The applicability of our mixed models were
assessed by examining models with and without the random effect of data collection site, the
linearity of CSF proteins over time within subjects and the normality of model residuals using
diagnostic plots. All models were tested in both AD patients (n = 59) and MCI subjects
(n = 142) with serial MMSE measurements. As a large number of proteins from the RBM panel
were tested we used a false discovery rate correction to account for multiple comparisons.
A multivariate support vector machine (SVM) algorithm was applied to the ADNI cohort,
in an unbiased fashion, to distinguish AD patients from CN individuals. In particular, a linear
SVM algorithm was constructed using the LIBSVM implementation . In the algorithm, the
parameter C (representing the error/trade off parameter used for adjusting separation error in
the creation of separation space) was optimized using 5-fold cross validation on the training
set. The grid search routine suggested by Hsu et al (2010)  was implemented to identify
optimal parameter settings for differentiating AD from CN individuals. A multi-kernel
learning approach for linear SVM  was implemented for treating variables of a different nature.
A general framework for kernel methods used to integrate data from different modalities has
been described previously in more extensive detail [36–38].
To identify a subset of CSF proteins associated with AD, we adopted a recursive feature
elimination (RFE) wrapper. The final subset of CSF proteins (CSF RFE subset) were then
combined with CSF biomarkers and regional MRI measurements to test their utility for disease
classification. Classification accuracy in each of these models was evaluated using ten-fold
cross validation. Measures of accuracy, sensitivity, specificity, and area under the curve (AUC)
were used to compare AD vs. CN models.
MCI subjects were divided into subjects that progressed to an AD diagnosis
(MCI-converters) and others that remained clinically stable over a 12 month follow up period (MCI
nonconverters). Subsequently, models from the AD vs. CN comparisons were used as training
classifiers to prospectively predict MCI to AD conversion in MCI converters (MCI-c), as well as
predicting MCI non-converters (MCI-nc) that remained stable at 12 months. Discriminant
scores from the model were then used to classify MCI subjects as either having an AD-like or
CN-like phenotype. The combined model (CSF RFE subset + CSF biomarkers + regional MRI
measures) was also used to predict MCI to AD conversion in moderately late MCI-c (subjects
that progressed to AD between 18–24 months follow up) and late MCI-c (subjects that
progressed to AD at 36 months). MCI-nc predictions were also made using the combined model
for subjects that remained clinically stable between 0–12 months, 18–24 months and 36
months follow up.
The R statistical software environment (v. 3.1.0; The R Foundation for Statistical
Computing), was used to perform all statistical analyses.
Baseline sample characteristics of the ADNI cohort for demographic, cognitive, MRI and CSF
biomarkers are presented by diagnostic group in Table 1. Significant differences between
groups were found in hippocampal and entorhinal volume, as well as, SPARE-AD score, CSF
biomarkers of AD, MMSE score and ApoE ε4 genotype. Subject age, gender and years of
education were not found to differ significantly between groups.
CSF proteins from the multiplex RBM panel associated with
neuroimaging markers of brain atrophy and CSF biomarkers of AD
Due to the exploratory nature of this study we first tested the association of the entire multiplex
panel of CSF proteins (n = 83) with neuroimaging and CSF biomarkers to identify candidates
related AD pathogenesis. Associations were tested using a partial spearman rank correlation
test that co-varied for the effects of age, gender, years of education and ApoE ε4 genotype. For
several proteins (n = 50) we found an association with either neuroimaging markers of brain
atrophy or CSF biomarkers of AD (Fig 1). Many proteins in this subset (n = 37) were also
found to be significantly associated with both P-tau181 and T-tau CSF levels. Reduced levels of
CgA were found to be significantly associated across all comparisons with neuroimaging and
CSF biomarkers, but only remained significant in association with hippocampal (p = <0.001)
and entorhinal volume (p = 0.008) after multiple comparison correction. Increased levels of
Fatty Acid Binding Protein (FABP) emerged as the most significantly associated with CSF
levels of P-tau181 and T-tau as well as SPARE-AD score. Results from our partial spearman rank
correlation test are displayed in S3 Table. Bear in mind most associations were mild and
reflected by p-values that were uncorrected for multiple comparisons.
CSF proteins associated with different ApoE gene polymorphisms
CSF proteins from the RBM panel were also tested in relation to the ApoE polymorphism
rather than diagnostic status. Significant differences in CSF levels were examined by ApoE
genotype (ε2 carriers, ε3 carriers and ε4 carriers). We found that 9 CSF proteins were
associated with the overall effect of ApoE genotype (Apo E, FABP, FGF-4, IL-8, AGRP, MIF, IL-3,
ANG-2, and Osteopontin). However, only CSF levels of Apo E, IL-3 and MIF were found to
differ between ApoE groups. CSF levels of these proteins compared to each ApoE group are
shown in Fig 2. In particular, the strongest overall effect was observed for CSF levels of Apo- E
which passed multiple comparison correction (p = .00046; FDR corrected = .034). Pairwise
comparisons revealed that Apo-E levels were significantly lower in ε4 carriers irrespective of
diagnosis compared to ε2 carriers (t = -3.63; p = < .0001) and significantly lower in ε3 carriers
compared to ε2 carriers (t = -2.57; p = .027).
CSF proteins related to the rate of cognitive decline on longitudinal
We also tested the association of baseline CSF proteins with the rate of cognitive decline using
at least three or four serial measurements of MMSE score. Firstly, we tested this in a sample of
AD (n = 65)
MCI (n = 142)
CN (n = 88)
Data are represented as mean ± and standard deviation. AD = Alzheimer’s disease, MCI = Mild Cognitive Impairment, CN = cognitively normal
individuals, MMSE = Mini Mental State Examination, ICV = Intracranial Volume, SPARE-AD score = Spatial Pattern of Abnormalities for Recognition of
Early AD. Chi-square was used for gender and ApoE ε4 genotype comparison. One way ANOVA with Bonferroni post hoc test was used for continuous
a indicates significant compared to the MCI group.
b indicates significant compared to the CN group.
Fig 1. Heatmap of baseline CSF proteins that were significantly associated with regional MRI measures, SPARE-AD score or CSF biomarkers in
AD patients and MCI subjects (n = 207).
Fig 2. CSF proteins significantly associated with different ApoE gene polymorphisms (ε2 carriers, ε3 carriers, and ε4 carriers). (A) CSF levels of
ApoE protein between ApoE groups; (B) CSF levels of Interleukin-3 (IL-3) between ApoE groups and (C) CSF levels of Macrophage migration inhibitory
factor (MIF) between ApoE groups. *These units refer to data before transformation.
AD patients (n = 59) and found no CSF proteins were able to significantly predict a
longitudinal change in MMSE score. However, in a sample of MCI subjects (n = 142) we found three
proteins (SGOT, MCP-1, and IL-6r) were able to significantly predict cognitive decline
(Table 2). Again, it should be noted that these associations were mild and no protein remained
significant after multiple comparison correction.
The recursive feature elimination (RFE) wrapper method identified a subset of 24 CSF proteins
which best distinguished AD patients from CN individuals (Table 3). Overall, we found that
the inclusion of these CSF proteins from the RBM panel improved the accuracy and specificity
of models. In particular, combining the CSF RFE subset with CSF biomarkers resulted in an
accuracy of 84.3% and an AUC of 91% (Table 4). We found that combining the CSF RFE
subset improved the sensitivity in a model generated using CSF biomarkers from 70.8% to 83.1%
Linear mixed effect models
Linear mixed effect model results are displayed as on b-coefficients (β), standard-error
(S.E) and P-values for the interaction terms between proteins and time (years from baseline).
Data were adjusted for age, gender, years of education and ApoE genotype as fixed effects and subject
code and site-id as random effects.
which was statistically significant (Venkatraman’s Test: Z = 2.94; p = .0042). Moreover, the
CSF RFE subset combined with CSF biomarkers and regional MRI measures, achieved an
accuracy of 91.5% (SEN = 87.7%, SPE = 94.3%, AUC = 0.95) which was significantly better than
using CSF biomarkers alone (Z = 2.91; p = .0036) (Fig 3). In the combined model (CSF RFE
subset + CSF biomarkers + regional MRI measures) we found CgA, FABP, MMP-2, and PPP
Data are percentages and confidence intervals are presented in parenthesis.
ACC = Accuracy, SENS = sensitivity, SPE = specificity, AUC = area under the curve.
The combined model includes regional MRI measures, CSF biomarkers of AD and the CSF RFE subset of
proteins (n = 24).
contributed most strongly toward the detection of AD. However, the regional MRI measures
and CSF biomarkers model gave the best result with an accuracy of 92.2% (SEN = 85.7%,
SPE = 96.4%, AUC = 0.96) but this was not found to be significantly better than the CSF RFE
subset combined with CSF biomarkers and regional MRI measures (Z = 0.38; p = 0.70). All
model results are shown in Table 4.
MCI to AD conversion prediction
Over a follow up period of 36 months, 72 MCI subjects (50.7%) from our sample converted to
an AD diagnosis. Table 5 shows the number of MCI subjects that were predicted as either
ADlike or CN-like at a 12 month follow-up interval using all AD vs. CN models.
Firstly, we tested all models in early MCI subjects who progressed to an AD diagnosis
between 0–12 months (n = 34) [Early MCI-c]. We found that the inclusion of the CSF RFE
subset with CSF biomarkers and regional MRI measures provided the best result, accurately
predicting 94.1% of MCI-c progressing to AD whereas the regional MRI measures and CSF
biomarker model was only able to achieve a prediction of 76.5%. Therefore, we further tested
the combined model in moderately late MCI-c (n = 26) who were also correctly predicted with
a 92.3% accuracy as progressing to AD, and late MCI-c (n = 12) who were predicted correctly
with an 82.4% accuracy. Fig 4a displays the predicted probabilities from the combined model
of MCI subjects that progressed to AD. For comparison, we also overlaid the predicted
probabilities of AD patients and CN individuals. The majority of subjects that converted to AD at
different follow up periods were found to already possess an AD-like phenotype at the
prodromal MCI stage (p > 0.05; Kolmogorov-Smirnov test).
In contrast, MCI-nc predictions were less accurate with predictions ranging from 57.4% to
25.0%. Over a 36 month follow up period 70 MCI subjects (49.3%) remained clinically stable.
The regional MRI measures model was found to yield the best prediction at a 12 month follow
up. For the combined model, Fig 4b displays an almost bimodal distribution of MCI-nc
predictive values, with some correctly predicted as CN-like and others predicted as having an
ADlike phenotype. Despite these subjects remaining clinically stable at their respective period of
follow up, some MCI subjects are expected to convert in the near future and as a result also
display an AD-like phenotype at baseline. Further follow ups will determine whether these
subjects remain clinically stable or convert to an AD diagnosis.
Fig 3. ROC curves from disease classification models for differentiating between AD and CN individuals.
MCI-c Classification (n = 34)
AD = Alzheimer’s disease, MCI = Mild Cognitive Impairment, MCI-c = MCI converter, MCI-nc = MCI
nonconverter, CN = Cognitively Normal.
*Sensitivity is the percentage of MCI-c subjects correctly classified as AD in bold.
**Specificity is the percentage of MCI-nc subjects correctly classified as CN in bold.
The combined model includes regional MRI measures, CSF biomarkers of AD and the CSF RFE subset of
proteins (n = 24).
Fig 4. Predictive values from the combined CSF RFE subset CSF biomarker and regional MRI measures model for MCI to AD conversion
prediction at several follow up timepoints. (A) Predictive values of MCI-c progressing to AD at different follow up timepoints overlaid with predictive values
of AD and CN individuals and (B) predictive values of MCI-nc at different follow up timepoints overlaid with predictive values of AD and CN individuals.
In this neuroimaging-proteomic study there were a number of key findings. Firstly, we
identified several CSF proteins (n = 50) related to neuroimaging phenotypes of brain atrophy and
CSF biomarkers of AD, suggesting that these candidates may be related to AD
pathophysiology. Second, a unique biochemical profile of CSF proteins was found to be associated with
ApoE genotype characterised by reduced levels of Apo E protein in ε4 carriers. Third, some
proteins (SGOT, MCP-1, and IL6-r) were found to be related to a longitudinal change in
MMSE score over a 4 year period. Although the statistical effects associated with this finding
were mild and no result passed multiple comparison correction, further studies will determine
whether they may serve as important prognostic factors related to the rate of cognitive decline.
More importantly, we showed that reducing the RBM panel to a subset of 24 CSF proteins
complemented existing AD biomarkers for AD detection and MCI to AD conversion
Our findings were in agreement with some previous studies identifying a panel of candidate
proteins associated with AD [15,21,39]. In particular, our first finding showed that several
proteins were associated with brain atrophy and CSF biomarkers, however, levels of CgA and
FABP emerged as the most consistently present across most our comparisons. Although the
effects associated with these findings were mild, previous studies have linked these candidates
to AD pathophysiology [40,41]. For instance, we found elevated levels of FABP protein to be
significantly related to neuroimaging SPARE-AD score which is in agreement with previous
findings reporting elevated levels of FABP protein in AD and prodromal MCI subjects
[15,21,42]. Increased levels of CgA protein related to hippocampal and entorhinal volume has
also been previously linked to early synaptic dysfunction in AD , reduced microglial
regulation of synaptic function , and Aβ1–42 metabolism in CN individuals .
We also found a unique biochemical profile of CSF proteins was associated with different
ApoE gene polymorphisms. For instance, ApoE protein and IL-3 levels were reduced in ε4
carriers, whilst MIF protein levels were elevated. Previous studies have also reported a peripheral
CSF signature associated with ApoE genotype  and similar findings have also been
observed in blood plasma . Moreover, many of the CSF candidates previously described in
the literature (e.g. FABP, FGF-4, IL-8, AGRP, ANG-2, and Osteopontin) also showed mild
associations with ApoE genotype, suggesting that the biological variability of proteins
identified in AD cases may also be in part driven by genotype status.
For differentiating between AD and CN individuals we found that a subset of proteins
(n = 24) from the RBM multiplex panel improved the accuracy and performance of models but
was unable to achieve a better accuracy when regional MRI measures and CSF biomarkers
were combined. Despite this, the combination of regional MRI measures, CSF biomarkers
(Aβ1–42, T-tau and P-tau) and the CSF RFE subset achieved an accuracy of 91.5%. In this
subset, four proteins namely CgA, FABP, MMP-2, and PPP were the strongest predictors for
distinguishing AD from CN individuals.
This is in agreement with previous studies showing that CSF candidates identified using
immunoassay panel technology can complement CSF biomarkers of AD for the earlier
detection of AD [21,46]. Recent neuroimaging-proteomic studies have also shown several proteins
to be associated with longitudinal rates of brain atrophy , as well as whole brain atrophy
. To our knowledge this is the first study to test whether CSF proteins from an
immunoassay panel can complement CSF biomarkers and regional MRI measures for disease
classification and prediction. Previous studies have suggested that the use of conventional imaging, such
as MRI, combined with biomarkers from different modalities may be complimentary to the
early and specific diagnosis of AD [48,49]. Several studies have reported that this combined
approach improved AD disease classification [10,36] and future MCI to AD conversion
For MCI to AD conversion prediction, the CSF RFE subset, CSF biomarker and regional
MRI measures model also gave the best results and outperformed all other models. The model
was particularly sensitive for correctly predicting MCI-c with an AD-like phenotype (Fig 4a)
suggesting that our panel of 24 proteins may also have the prognostic potential to detect
prodromal AD. However, the model failed to correctly detect MCI-nc, with a large proportion of
subjects being predicted as AD-like, suggesting that the model lacked specificity. Several
previous studies on MCI to AD conversion prediction have also noted the heterogeneity of the
MCI-nc group using similar high dimensional pattern classification algorithms [10,51]. It is
anticipated that many MCI-nc will convert to AD in the near future. Although future studies
with longer follow up times will refine our estimates of specificity, the ability to detect MCI-nc
many years prior to clinical diagnosis could provide useful tools for an earlier diagnosis.
Despite some promising results, there exist a number of limitations to our findings. Firstly,
although we identified a number of CSF proteins showing promise in AD detection and MCI
to AD conversion prediction our results are somewhat limited by the inability to validate these
candidates in an independent cohort. Therefore future studies are warranted to further explore
the prognostic potential of the candidates identified here in other well-characterised
prospective cohorts. Nonetheless, we do show that the panel of CSF proteins for detecting AD also
have a good prognostic potential for detecting AD in the prodromal or amnestic MCI stage.
Secondly, CSF proteins identified in this study were from a multiplex panel of proteins known
to be associated with microglial activity and synaptic function. It may be likely that an
alternative set of CSF proteins unrelated to these processes could also show strong effects in detecting
AD and predicting MCI to AD conversion.
In summary, the relation of CSF proteins to key neuroimaging phenotypes and traditional
CSF biomarkers provides some evidence of their importance in reflecting early
neuropathological changes in AD pathogenesis. Combining a subset of proteins (n = 24) from the RBM
multiplex panel with established biomarkers in AD provides further evidence to implicate the role of
peripheral CSF proteins for improving the accuracy and prognostic ability of biomarkers for
disease classification and progression. Future studies are warranted to further validate our
findings with more generalisability in other well-characterised independent cohorts.
S3 Table. Baseline CSF proteins that were significantly associated with regional MRI
measures, SPARE-AD score or CSF biomarkers in AD patients and MCI subjects (n = 207).
Data used in the preparation of this article were obtained from the ADNI database (adni.loni.
ucla.edu). As such, the investigators within the ADNI study contributed to the design and
implementation of ADNI or provided data but did not participate in the analysis or writing of
Data collection and sharing for this project were funded by the ADNI (grant U01
AG024904 from the National Institutes of Health). ADNI is funded by the National Institute
on Aging and the National Institute of Biomedical Imaging and Bioengineering, as well as
through generous contributions from the following: Abbott, Alzheimer’s Association,
Alzheimer’s Drug Discovery Foundation, Amorfix Life Sciences Ltd, AstraZeneca, Bayer HealthCare,
BioClinica Inc, Biogen Idec Inc, Bristol-Myers Squibb Company, Eisai Inc, Elan
Pharmaceuticals Inc, Eli Lilly and Company, F. Hoffmann-La Roche Ltd and its affiliated company
Genentech Inc, GE Healthcare, Innogenetics NV, Janssen Alzheimer Immunotherapy, Johnson &
Johnson Pharmaceutical Research & Development LLC, Medpace Inc, Merck & Co, Meso
Scale Diagnostics LLC, Novartis Pharmaceuticals Corporation, Pfizer Inc, Servier, Synarc, and
Takeda Pharmaceutical Company. The Canadian Institutes of Health Research is providing
funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by
the Foundation for the National Institutes of Health (http://www.fnih.org). The grantee
organization is the Northern California Institute for Research and Education, and the study is
coordinated by the Alzheimer’s Disease Cooperative Study at the University of California, San Diego.
ADNI data are disseminated by the Laboratory of Neuro Imaging at the University of
California, Los Angeles. This research was also supported by grants P30 AG010129 and K01
AG030514 from the National Institutes of Health and by the Dana Foundation.
This study was supported by funds from the NIHR Biomedical Research Centre for Mental
Health and NIHR Biomedical Research Unit for Dementia at the South London and Maudsley
NHS Foundation Trust and Institute of Psychiatry, Kings College London and Alzheimer’s
Conceived and designed the experiments: WK CA SJK EW AS VG RD MS SL. Analyzed the
data: WK CA MS. Contributed reagents/materials/analysis tools: WK OD MT SN SM MS.
Wrote the paper: WK EW AS.
23. ADNI cerebrospinal fliud aliquot inventory description report [Internet]. Available:
24. ADNI clinical procedures manual [Internet]. Dec. 10.1093/infdis/jis910
1. Jack CR , Knopman DS , Jagust WJ , Shaw LM , Aisen PS , Weiner MW , et al. Hypothetical model of dynamic biomarkers of the Alzheimer's pathological cascade . Lancet Neurol . 2010 ; 9 : 119 - 28 . PMID: 20083042
2. Villemagne VL , Burnham S , Bourgeat P , Brown B , Ellis K a , Salvado O , et al. Amyloid β deposition, neurodegeneration, and cognitive decline in sporadic Alzheimer's disease: a prospective cohort study . Lancet Neurol . 2013 ; 12 : 357 - 67 . PMID: 23477989
3. Braak H , Braak E. Neuropathological stageing of Alzheimer-related changes . Acta Neuropathol . Springer; 1991 ; 82 : 239 - 259 .
4. Jack CR , Wiste HJ , Weigand SD , Rocca WA , Knopman DS , Mielke MM , et al. Age-specific population frequencies of cerebral β-amyloidosis and neurodegeneration among people with normal cognitive function aged 50-89 years: a cross-sectional study . Lancet Neurol . 2014 ; 13 : 997 - 1005 . PMID: 25201514
5. Fox NC , Warrington EK , Freeborough PA , Hartikainen P , Kennedy AM , Stevens JM , et al. Presymptomatic hippocampal atrophy in Alzheimer's disease . A longitudinal MRI study . Brain. OXFORD UNIV PRESS ; 1996 ; 119 : 2001 - 2007 .
6. Desikan RS , Cabral HJ , Hess CP , Dillon WP , Glastonbury CM , Weiner MW , et al. Automated MRI measures identify individuals with mild cognitive impairment and Alzheimer's disease . Brain. Oxford University Press ; 2009 ; 132 : 2048 - 2057 .
7. Blennow K , Hampel H. CSF markers for incipient Alzheimer's disease . Lancet . 2003 ; 2 : 605 - 613 . PMID: 14505582
8. Shaw LM , Korecka M , Clark CM , Lee VM-Y , Trojanowski JQ . Biomarkers of neurodegeneration for diagnosis and monitoring therapeutics . Nat Rev Drug Discov . 2007 ; 6 : 295 - 303 . PMID: 17347655
9. Davatzikos C , Bhatt P , Shaw LM , Batmanghelich KN , Trojanowski JQ . Prediction of MCI to AD conversion, via MRI , CSF biomarkers, and pattern classification . Neurobiol Aging. Elsevier Inc .; 2010 ;
10. Westman E , Muehlboeck J-S , Simmons A. Combining MRI and CSF measures for classification of Alzheimer's disease and prediction of mild cognitive impairment conversion . Neuroimage. Elsevier B.V.; 2012 ; 62 : 229 - 38 .
11. Mattsson N , Zetterberg H , Hansson O , Andreasen N , Parnetti L , Jonsson M , et al. CSF biomarkers and incipient Alzheimer disease in patients with mild cognitive impairment . JAMA . 2009 ; 302 : 385 - 93 . doi: 10.1001/jama.2009.1064 PMID: 19622817
12. Fagan AM , Perrin RJ . Upcoming candidate cerebrospinal fluid biomarkers of Alzheimer's disease . Biomark Med . 2012 ; 6 : 455 - 476 . doi: 10.2217/bmm. 12.42 PMID: 22917147
13. Guerreiro R , Wojtas A , Bras J , Carrasquillo M , Rogaeva E , Majounie E , et al. TREM2 variants in Alzheimer's disease . N Engl J Med . 2013 ; 368 : 117 - 27 . doi: 10.1056/NEJMoa1211851 PMID: 23150934
14. Zhang J , Goodlett DR , Peskind ER , Quinn JF , Zhou Y , Wang Q , et al. Quantitative proteomic analysis of age-related changes in human cerebrospinal fluid . Neurobiol Aging . 2005 ; 26 : 207 - 27 . PMID: 15582749
15. Hu WT , Chen-Plotkin A , Arnold SE , Grossman M , Clark CM , Shaw LM , et al. Novel CSF biomarkers for Alzheimer's disease and mild cognitive impairment . Acta Neuropathol . 2010 ; 119 : 669 - 78 . doi: 10. 1007/s00401- 010 - 0667 - 0 PMID: 20232070
16. Craig-Schapiro R , Perrin RJ , Roe CM , Xiong C , Carter D , Cairns NJ , et al. YKL-40: a novel prognostic fluid biomarker for preclinical Alzheimer's disease . Biol Psychiatry. Elsevier Inc .; 2010 ; 68 : 903 - 12 .
17. Weiner MW , Aisen PS , Jack CR , Jagust WJ , Trojanowski JQ , Shaw L , et al. The Alzheimer's disease neuroimaging initiative: progress report and future plans . Alzheimers Dement. Elsevier Ltd ; 2010 ; 6: 202 - 11 .
18. Shaw LM , Vanderstichele H , Knapik-Czajka M , Clark CM , Aisen PS , Petersen RC , et al. Cerebrospinal fluid biomarker signature in Alzheimer's disease neuroimaging initiative subjects . Ann Neurol . 2009 ; 65 : 403 - 413 . doi: 10.1002/ana.21610 PMID: 19296504
19. Shaw LM , Vanderstichele H , Knapik-Czajka M , Figurski M , Coart E , Blennow K , et al. Qualification of the analytical and clinical performance of CSF biomarker analyses in ADNI . Acta Neuropathol . 2011 ; 121 : 597 - 609 . doi: 10.1007/s00401- 011 - 0808 - 0 PMID: 21311900
20. Trojanowski JQ , Vandeerstichele H , Korecka M , Clark CM , Aisen PS , Petersen RC , et al. Update on the biomarker core of the Alzheimer's Disease Neuroimaging Initiative subjects . Alzheimer's Dement . Elsevier Ltd; 2010 ; 6: 230 - 238 .
21. Craig-Schapiro R , Kuhn M , Xiong C , Pickering EH , Liu J , Misko TP , et al. Multiplexed immunoassay panel identifies novel CSF biomarkers for Alzheimer's disease diagnosis and prognosis . PLoS One . 2011 ; 6 : e18850. doi: 10.1371/journal. pone.0018850 PMID: 21526197
22. Hu WT , Holtzman DM , Fagan AM , Shaw LM , Perrin R , Arnold SE , et al. Plasma multianalyte profiling in mild cognitive impairment and Alzheimer disease . Neurology . 2012 ; 79 : 897 - 905 . PMID: 22855860
25. ADNI: Alzheimer's Disease Neuroimaging Initiative [Internet] . Available: http://www.adni-info.org/ Scientists/ADNIStudyProcedures.aspx
26. Jack CR , Bernstein MA , Fox NC , Thompson P , Alexander G , Harvey D , et al. The Alzheimer's disease neuroimaging initiative (ADNI): MRI methods . J Magn Reson Imaging . Wiley Online Library ; 2008 ; 27 : 685 - 691 .
27. Simmons A , Westman E , Muehlboeck S , Mecocci P , Vellas B , Tsolaki M , et al. MRI measures of Alzheimer's disease and the AddNeuroMed study . Ann N Y Acad Sci . 2009 ; 1180 : 47 - 55 .
28. Simmons A , Westman E , Muehlboeck S , Mecocci P , Vellas B , Tsolaki M , et al. The AddNeuroMed framework for multi-centre MRI assessment of Alzheimer's disease: experience from the first 24 months . Int J Geriatr Psychiatry . 2011 ; 26 : 75 - 82 . doi: 10.1002/gps.2491 PMID: 21157852
29. Westman E , Simmons A , Zhang Y , Muehlboeck J-S , Tunnard C , Liu Y , et al. Multivariate analysis of MRI data for Alzheimer's disease, mild cognitive impairment and healthy controls . Neuroimage. Elsevier Inc .; 2011 ; 54 : 1178 - 87 .
30. Westman E , Simmons A , Muehlboeck J-S , Mecocci P , Vellas B , Tsolaki M , et al. AddNeuroMed and ADNI: similar patterns of Alzheimer's atrophy and automated MRI classification accuracy in Europe and North America . Neuroimage. Elsevier Inc.; 2011 ; 58 : 818 - 28 .
31. Westman E , Aguilar C , Muehlboeck J-S , Simmons A. Regional magnetic resonance imaging measures for multivariate analysis in Alzheimer's disease and mild cognitive impairment . Brain Topogr . 2013 ; 26 : 9 - 23 . doi: 10.1007/s10548- 012 - 0246 -x PMID: 22890700
32. Fan Y , Batmanghelich N , Clark CM , Davatzikos C. Spatial patterns of brain atrophy in MCI patients, identified via high-dimensional pattern classification, predict subsequent cognitive decline . Neuroimage . 2008 ; 39 : 1731 - 1743 . PMID: 18053747
33. Davatzikos C , Xu F , An Y , Fan Y , Resnick SM . Longitudinal progression of Alzheimer's-like patterns of atrophy in normal older adults: the SPARE-AD index . Brain . 2009 ; 132 : 2026 - 35 . doi: 10.1093/brain/ awp091 PMID: 19416949
34. Chang C , Lin C-J. LIBSVM : A Library for Support Vector Machines . ACM Trans Intell Syst Technol . 2011 ; 2 : 1 - 27 .
35. Hsu C , Chang C , Lin C . A Practical Guide to Support Vector Classification. Bioinformatics. Citeseer; 2010 ; 1 : 1 - 16 .
36. Zhang D , Wang Y , Zhou L , Yuan H , Shen D. Multimodal classification of Alzheimer's disease and mild cognitive impairment . Neuroimage. Elsevier Inc .; 2011 ; 55 : 856 - 867 .
37. Schölkopf B , Smola AJ . Learning with Kernels . Dietterich T, editor. Kybernetik . MIT Press; 2002 .
38. Aguilar C , Westman E , Muehlboeck J-S , Mecocci P , Vellas B , Tsolaki M , et al. Different multivariate techniques for automated classification of MRI data in Alzheimer's disease and mild cognitive impairment . Psychiatry Res. Elsevier ; 2013 ; 212 : 89 - 98 .
39. Mattsson N , Insel P , Nosheny R , Trojanowski JQ , Shaw LM , Jack CR , et al. Effects of cerebrospinal fluid proteins on brain atrophy rates in cognitively healthy older adults . Neurobiol Aging. Elsevier Ltd ; 2014 ; 35 : 614 - 22 .
40. Blennow K , Davidsson P , Wallin A , Ekman R. Chromogranin A in cerebrospinal fluid: a biochemical marker for synaptic degeneration in Alzheimer's disease ? Dementia . 1995 ; 6 : 306 - 311 . PMID: 8563783
41. Chiasserini D , Parnetti L , Andreasson U , Zetterberg H , Giannandrea D , Calabresi P , et al. CSF levels of heart fatty acid binding protein are altered during early phases of Alzheimer's disease . J Alzheimers Dis . 2010 ; 22 : 1281 - 8 . doi: 10.3233/ JAD-2010-101293 PMID: 20930282
42. Olsson B , Hertze J , Ohlsson M , Nägga K , Höglund K , Basun H , et al. Cerebrospinal fluid levels of heart fatty acid binding protein are elevated prodromally in Alzheimer's disease and vascular dementia . J Alzheimers Dis . 2013 ; 34 : 673 - 9 . doi: 10.3233/ JAD-121384 PMID: 23254629
43. Taylor DL , Diemel LT , Cuzner ML , Pocock JM . Activation of group II metabotropic glutamate receptors underlies microglial reactivity and neurotoxicity following stimulation with chromogranin A, a peptide up-regulated in Alzheimer's disease . J Neurochem . 2002 ; 82 : 1179 - 1191 . PMID: 12358765
44. Mattsson N , Insel P , Nosheny R , Zetterberg H , Trojanowski JQ , Shaw LM , et al. CSF protein biomarkers predicting longitudinal reduction of CSF β-amyloid42 in cognitively healthy elders . Transl Psychiatry . 2013 ; 3 : e293. doi: 10.1038/tp.2013.69 PMID: 23962923
45. Soares HD , Potter WZ , Pickering E , Kuhn M , Immermann FW , Shera DM , et al. Plasma Biomarkers Associated With the Apolipoprotein E Genotype and Alzheimer Disease. Arch Neurol . 2012 ; 1 - 8 .
46. Perrin RJ , Craig-Schapiro R , Malone JP , Shah AR , Gilmore P , Davis AE , et al. Identification and validation of novel cerebrospinal fluid biomarkers for staging early Alzheimer's disease . PLoS One . 2011 ; 6 : e16032. doi: 10.1371/journal. pone.0016032 PMID: 21264269
47. Paterson RW , Bartlett JW , Blennow K , Fox NC , Shaw LM , Trojanowski JQ , et al. Cerebrospinal fluid markers including trefoil factor 3 are associated with neurodegeneration in amyloid-positive individuals . Transl Psychiatry . Nature Publishing Group ; 2014 ; 4: 419 .
48. Vemuri P , Wiste HJ , Weigand SD , Shaw LM , Trojanowski JQ , Weiner MW , et al. MRI and CSF biomarkers in normal, MCI, and AD subjects: predicting future clinical change . Neurology. 28 Jul 2009 73: 294 - 301 . doi: 10.1212/WNL.0b013e3181af79fb PMID: 19636049
49. Apostolova LG , Hwang KS , Andrawis JP , Green AE , Babakchanian S , Morra JH , et al. 3D PIB and CSF biomarker associations with hippocampal atrophy in ADNI subjects . Neurobiol Aging. Elsevier Inc .; 2010 ; 31 : 1284 - 1303 .
50. Cui Y , Liu B , Luo S , Zhen X , Fan M , Liu T , et al. Identification of Conversion from Mild Cognitive Impairment to Alzheimer 's Disease Using Multivariate Predictors. PLoS One. Public Library of Science ; 2011 ; 6: e21896 .
51. Misra C , Fan Y , Davatzikos C. Baseline and longitudinal patterns of brain atrophy in MCI patients, and their use in prediction of short-term conversion to AD: results from ADNI . Neuroimage. Elsevier Inc.; 2009 ; 44 : 1415 - 22 .