Journal of Alzheimer’s Disease 74 (2020) 213–225 213 DOI 10.3233/JAD-190434 IOS Press Validation of Plasma Proteomic Biomarkers Relating to Brain Amyloid Burden in the EMIF-Alzheimer’s Disease Multimodal Biomarker Discovery Cohort Sarah Westwooda , Alison L. Bairda , Sneha N. Ananda , Alejo J. Nevado-Holgadoa , Andrey Kormilitzina , Liu Shia , Abdul Hyeb , Nicholas J. Ashtonb,c,d , Angharad R. Morgane , Isabelle Bosf , Stephanie J.B. Vosf , Susan Bakernn , Noel J. Buckleya , Mara Ten Kateg , Philip Scheltensg , Charlotte E. Teunisseng , Rik Vandenbergheh , Silvy Gabelh,i , Karen Meersmansh,i , Sebastiaan Engelborghsj,k , Ellen E. De Roeckj,k , Kristel Sleegersk,l , Giovanni B. Frisonim,n , Olivier Blino , Jill C. Richardsonp , Régis Bordetq , José L. Molinuevor,s , Lorena Ramis , Anders Wallint , Petronella Kettunent , Magda Tsolakiu , Frans Verheyf , Alberto Lléov , Isabel Salaw , Julius Poppx,y , Gwendoline Peyratoutx , Pablo Martinez-Lagez , Mikel Taintaz , Peter Johannsenaa , Yvonne Freund-Levibb,b , Lutz Frölichcc , Valerija Dobricicdd , Cristina Legido-Quigleyee,ff , Lars Bertramdd,gg , Frederik Barkhofhh,ii , Henrik Zetterbergc,jj,kk,ll , B. Paul Morgane , Johannes Strefferk,mm , Pieter Jelle Visserf,g and Simon Lovestonea,pp,∗ a Department of Psychiatry, University of Oxford, UK b Maurice Wohl Clinical Neuroscience, Institute of Psychiatry, Psychology and Neuroscience, Kings College London, London, UK c Department of Psychiatry and Neurochemistry, University of Gothenburg, Mölndal, Sweden d Wallenberg Centre for Molecular & Translational Medicine, University of Gothenburg, Gothenburg, Sweden e Dementia Research Institute Cardiff, Cardiff University, Cardiff, UK f Department of Psychiatry and Neuropsychology, School for Mental Health and Neuroscience, Alzheimer Centrum Limburg, Maastricht University, Maastricht, the Netherlands g Alzheimer Center, VU University Medical Center, Amsterdam, the Netherlands h University Hospital Leuven, Leuven, Belgium i Laboratory for Cognitive Neurology, Department of Neurosciences, KU Leuven, Belgium j Department of Neurology and Memory Clinic, Hospital Network Antwerp (ZNA) Middelheim and Hoge Beuken, University of Antwerp, Antwerp, Belgium k Center for Neurosciences, Vrije Universiteit Brussel (VUB), Brussels, Belgium l Neurodegenerative Brain Diseases Group, Center for Molecular Neurology, VIB, Belgium m University of Geneva, Geneva, Switzerland n IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli, Brescia, Italy o AIX Marseille University, INS, Ap-Hm, Marseille, France p Neurosciences Therapeutic Area, GlaxoSmithKline R&D, Stevenage, UK q University of Lille, Inserm, CHU Lille, France r Alzheimer’s Disease & Other Cognitive Disorders Unit, Hopsital Clínic-IDIBAPS, Barcelona, Spain s Barcelona Beta Brain Research Center, Unversitat Pompeu Fabra, Barcelona, Spain ∗ Correspondence to: Professor Simon Lovestone, Department of Psychiatry, University of Oxford, UK. E-mail: simon.love stone@psych.ox.ac.uk. ISSN 1387-2877/20/$35.00 © 2020 – IOS Press and the authors. All rights reserved This article is published online with Open Access and distributed under the terms of the Creative Commons Attribution Non-Commercial License (CC BY-NC 4.0). 214 S. Westwood et al. / Validation of Plasma Biomarkers of Aβ t Institute of Neuroscience and Physiology, Sahlgrenska Academy at University of Gothenburg, Gothenburg, Sweden u 1st Department of Neurology, AHEPA University Hospital, Makedonia, Thessaloniki, Greece v Hospital de la Santa Creu i Sant Pau, Barcelona, Spain w Department of Neurology, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain x University Hospital of Lausanne, Lausanne, Switzerland y Geriatric Psychiatry, Department of Mental Health and Psychiatry, Geneva University Hospitals, Geneva, Switzerland z CITA-Alzheimer Foundation, San Sebastian, Spain aa Danish Dementia Research Centre, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark bb Department of Neurobiology, Caring Sciences and Society (NVS), Division of Clinical Geriatrics, Karolinska Institutet, and Department of Geriatric Medicine, Karolinska University Hospital Huddinge, Stockholm, Sweden cc Department of Geriatric Psychiatry, Zentralinstitut für Seelische Gesundheit, University of Heidelberg, Mannheim, Germany dd Lübeck Interdisciplinary Platform for Genome Analytics, Institutes of Neurogenetics and Cardiogenetics, University of Lübeck, Lübeck, Germany ee Kings College London, London, UK ff The Systems Medicine Group, Steno Diabetes Center, Gentofte, Denmark gg Department of Psychology, University of Oslo, Oslo, Norway hh Department of Radiology and Nuclear Medicine, VU University Medical Center, Amsterdam, The Netherland ii UCL Institutes of Neurology and Healthcare Engineering, London, UK jj Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden kk UK Dementia Research Institute at UCL, London, UK ll Department of Neurodegenerative Disease, UCL Institute of Neurology, London, UK mm UCB, Braine-l’Alleud, Belgium, formerly Janssen R&D, LLC. Beerse, Belgium at the Time of Study Conduct nn Janssen R&D, Titusville, NJ, USA pp Janssen R&D, UK formerly affiliation (1) at the Time of the Study Conduct Handling Associate Editor: Robert Rissman Accepted 27 December 2019 Abstract. We have previously investigated, discovered, and replicated plasma protein biomarkers for use to triage potential trials participants for PET or cerebrospinal fluid measures of Alzheimer’s disease (AD) pathology. This study sought to undertake validation of these candidate plasma biomarkers in a large, multi-center sample collection. Targeted plasma analyses of 34 proteins with prior evidence for prediction of in vivo pathology were conducted in up to 1,000 samples from cognitively healthy elderly individuals, people with mild cognitive impairment, and in patients with AD-type dementia, selected from the EMIF-AD catalogue. Proteins were measured using Luminex xMAP, ELISA, and Meso Scale Discovery assays. Seven proteins replicated in their ability to predict in vivo amyloid pathology. These proteins form a biomarker panel that, along with age, could significantly discriminate between individuals with high and low amyloid pathology with an area under the curve of 0.74. The performance of this biomarker panel remained consistent when tested in apolipoprotein E ␧4 non-carrier individuals only. This blood-based panel is biologically relevant, measurable using practical immunocapture arrays, and could significantly reduce the cost incurred to clinical trials through screen failure. Keywords: Alzheimer’s disease, amyloid-␤, biomarkers, plasma, proteomics INTRODUCTION prodromal and pre-symptomatic stages in the belief that disease modification efforts are most likely to Clinical trials for Alzheimer’s disease (AD) mod- be effective earlier in the disease process [1, 2]. ification have recently started to target the earlier However, conducting trials in prodromal/preclinical S. Westwood et al. / Validation of Plasma Biomarkers of Aβ 215 individuals necessitates the use of biomarkers to ‘cases’ from cognitively healthy elderly ‘controls’ detect evidence of in vivo AD pathology. Currently, [5]. However, the relatively low rate of replication pathology detection is possible using biochemical of individual biomarkers across studies may be in measures in cerebrospinal fluid (CSF) obtained from part due to issues of a study design that compares lumbar puncture and by positron emission tomogra- people with clinical AD, some of whom will not phy (PET) coupled to ligands for amyloid imaging. have pathology, to cognitively healthy elderly con- Both methods are employed routinely in clinical stud- trols, many of whom will harbor silent pathology. ies but their limitations impact significantly upon the Other factors limiting replication include technical efficiency of clinical trials. Both CSF sampling and issues such as assay variance and quality, differences PET can be costly and invasive and are therefore not in sampling and storage protocols and the frequently suitable for large-scale screening or where repeated small size of many studies [6, 7]. Nonetheless, such measures are desirable. The relatively low prevalence studies demonstrate that there is a signature of dis- of amyloid pathology in people with mild cognitive ease detectable in blood and the task is now to find a impairment (MCI), and even more so in cognitively signature that is reproducible. healthy individuals, inevitably results in high screen To address this, we have gradually refined our failure rates when trying to detect prodromal or pre- study design, focusing on an ‘endophenotype’ clinical AD [3]. The cost of this screen failure by approach predicating on outcomes not of clinical either CSF or PET can be prohibitive as well as con- diagnosis, but of a phenotype indicative of dis- tributing to the delay in start-up of clinical trials. ease such as brain atrophy measured by structural More worryingly, given that such screening is likely MRI or amyloid-␤ (A␤) plaque burden measured to be mandatory as part of clinical implementation by PET and CSF. Using this approach, we have of a successful therapeutic, this screen failure rate is identified putative plasma markers relating to in likely to constitute an obstacle to clinical translation vivo AD pathology and disease progression using a through a combination of cost factors and the capacity range of proteomic approaches including mass spec- of health systems to enable lumbar puncture or PET trometry, SOMAscan, and immunocapture [8–15]. imaging at the scale likely to match demands of their In addition, we also performed a series of iterative populations. One option to overcome this issue is to studies targeting complement and related inflam- employ prediction methods such as apolipoprotein matory proteins; neuroinflammation itself being an E (APOE) genotyping prior to trial entry, as APOE endophenotype associated with disease and amyloid ␧4 carriers are more likely to harbor AD pathology. pathology [16]. Many of the proteins identified in However, not all individuals with preclinical or pro- these discovery-phase ‘endophenotype’ studies repli- dromal AD have an APOE ␧4 allele and furthermore cate across proteomic platforms and different cohort APOE genotyping only reveals risk and does not indi- types. However, unsurprisingly given the wide range cate current pathological state. Therefore, in order to of techniques and study designs, not every protein increase the efficiency of recruitment to clinical tri- identified replicates in every study. The aim of the als, a cost effective, minimally invasive method that present study was to determine the most replica- can be implemented on a large scale to predict cur- ble set of plasma protein markers predicting in vivo rent AD pathology would be enormously valuable. brain amyloid pathology, by testing 31 of our pre- A blood-based assay that predicted likely pathologi- viously discovered candidate biomarkers in a large, cal load would become part of a diagnostic screening multi-center cohort of individuals with high and low funnel triaging potential trials participants or users of amyloid burden. therapy to direct markers of pathology such as CSF measures or PET imaging. METHODS Over a decade ago, we conducted a large agnos- tic, or untargeted, proteome-wide, discovery study of Subjects: EMIF-AD Multimodal Biomarker blood biomarkers in dementia [4]. This study was Discovery study (EMIF-AD MBD) cohort successful in detecting a signal in the blood that reflected the presence of AD-type dementia and since The EMIF-AD MBD is part of the European then many other studies, by others and by ourselves, Medical Information Framework for Alzheimer’s have aimed to replicate and improve upon this orig- Disease (EMIF-AD; http://www.emif.eu/), a Euro- inal signal and have identified blood-based protein pean wide collaboration to facilitate the re-use of biomarkers able to distinguish AD-type dementia existing healthcare data and the sharing of cohort 216 S. Westwood et al. / Validation of Plasma Biomarkers of Aβ samples for the benefit of AD research. The EMIF- Luminex xMAP (Cat#: HNDG1MAG-36K-06, AD MBD study design, including subject selection HNDG2MAG-36K-05, HNDG3MAG-36K-07, HN criteria, clinical diagnoses, brain amyloid classifi- D2MAG-39K-02, HKI6MAG-99K-03, HNDG1 cation, and plasma sample collection have all been MAG-36K-01), ELISA (Cat#: CSB-EL008551HU described previously [17]. In essence though, we and CSB-E13319H), and MSD assays (in-house sought to assemble a collection of samples from par- optimized using U-plex platform). All assays were ticipants in cohort studies from across the full clinical performed according to the manufacturer’s instruc- disease spectrum from preclinical through prodro- tions. Overall, concentrations of 25 proteins were mal to advanced disease, in each category seeking determined at Oxford while 6 proteins were analyzed to balance those with amyloid pathology to those at Cardiff. without. To do this we selected, using existing data, samples for inclusion from apparently normal, cog- Brain amyloid measurements and group nitively healthy elderly controls, from participants classifications with diagnosed MCI and from people with estab- lished AD-type dementia. Samples were selected Measurement and classification of in vivo amy- within each category with proven high and low loid burden in the EMIF-AD MBD cohort has been amyloid load as previously described [17]. Overall, described previously [17]. Briefly, where CSF was 1,221 participants were recruited to the EMIF-AD available, A␤1-40 and A␤1-42 were measured using MBD study from 11 European cohorts. Each par- the V-PLEX Plus A␤ Peptide Panel 1 (6E10) Kit ent cohort was approved by the local medical ethics from Meso Scale Discovery in a central laboratory committee. (Gothenburg University, Sweden) and the A␤42/40 The present study selected two sub-cohorts of par- ratio was established. The A␤ group was defined ticipants from the EMIF-AD MBD study, all with by the A␤42/40 ratio, using a cut-off of <0.063 to plasma samples available for analysis. Firstly, 1,000 determine abnormality. Where CSF samples were individuals comprising 408 cognitively healthy indi- unavailable, then the locally measured CSF A␤42 viduals, 400 individuals with MCI, and 192 AD-type value and cut-off point provided by the parent cohort dementia patients were included for proteomic analy- or the standardized uptake value ratio (SUVR) and sis in the University of Oxford laboratories. Secondly, cut-off point from an amyloid PET scan was used. 866 individuals (93 AD, 413 MCI, 360 control) were The above measurements were combined into a included for proteomic analysis in the laboratories at continuous variable using z-scoring. The A␤ Z-score Cardiff University. 679 (78%) of the Cardiff sam- was calculated using the mean and standard devia- ple set were also included in the Oxford sample tion of the control subjects as a reference. In cases set. Participants were included in these ‘Oxford’ and where an individual had multiple measures of amy- ‘Cardiff’ sample sets from across three multi-center loid (e.g., CSF and PET), all data available were studies: DESCRIPA [18], EDAR [19], and Pharma- used to generate the mean and standard deviation Cog [20], and eight single center studies: Amsterdam for each measure. However, the measure included in [21], Antwerp [22], San Sebastian GAP [23], Gothen- the final A␤ Z-score was selected from each individ- burg [24], Barcelona IDIBAPS [25], Lausanne [26], ual in the following order of priority: CSF A␤42/40 Leuven [27], and Barcelona St Pau [28]. Sample num- ratio, local CSF A␤42 , or the amyloid PET SUVR. ber differences between the ‘Oxford’ and ‘Cardiff’ PET amyloid Z-scores were multiplied by –1 in sample sets were necessary due to plasma sample order to be combined with CSF derived amyloid availability. Differences in the distribution of partic- Z-scores. ipants across multiple diagnostic categories between the ‘Oxford’ and ‘Cardiff’ sample sets are displayed in Supplementary Figure 1. The methods for address- CSF tau measurements ing the challenge these differences presented are described in full in Bos et al. [17]. To assess in vivo tau pathology, continuous phos- phorylated tau (p-tau) and total tau (t-tau) values were Plasma analyses obtained from the parent cohorts [17]. As sites were not standardized to each other, the p-tau and t-tau val- Targeted plasma protein analyses were conducted ues were Z-scored with controls within each data set at both Oxford and Cardiff laboratories using as a reference. S. Westwood et al. / Validation of Plasma Biomarkers of Aβ 217 MRI data or non-carriers according to their genotype status at rs429358 (C-allele = ␧4). Missing values for APOE Full details on the MRI data acquisition, visual rat- ␧4 status were imputed using regression analyses ing check, and region of interest measurements have within study, based on significant predictors (age, been previously reported [17]. Briefly, TI-weighted gender, MMSE, cognitive scores) [17]. images, acquired according to local protocols, were collected from each site, each image was visually Statistical analyses assessed and Freesurfer used to obtain volumetric measurements. All statistical analyses were completed using R (version 3.3.2) and all statistical tests were two-tailed. Clinical and cognitive data Individual participant data was excluded where there was a long interval (>1 year), or missing data on the Baseline clinical information and neuropsycho- time interval, between plasma collection and mea- logical test scores were collected from each local surement of the outcome variable (amyloid status, site, harmonized, pooled and stored in an online data n = 69 and 30 excluded from Oxford and Cardiff sam- platform using tranSMART [29]. Full details of the ple sets, respectively; MMSE, n = 73 and 39 excluded clinical information provided by each site and the from Oxford and Cardiff, respectively; brain vol- harmonization process has been previously described ume, n = 121 and 97 excluded from Oxford and [17]. Cardiff, respectively). Baseline cohort characteristics In all cohorts, a diagnosis of AD-type demen- between high and low amyloid groups were compared tia was made according to the National Institute using Mann-Whitney U test. All regression analy- of Neurological and Communicative Disorders and ses included age as a covariate. p-values and false Stroke-Alzheimer’s Disease and Related Disorders discovery rate corrected q-values are reported. Association (NINCDS-ADRDA) criteria [30]. The Petersen criteria [31] was used to diagnose MCI in Univariate statistics nine cohorts, while two cohorts used the Winblad Univariate statistics were performed using iden- et al. criteria [32] for this diagnosis. Normal perfor- tical statistical methods for both the ‘Oxford’ and mance on neuropsychological assessment (1.5 SD of ‘Cardiff’ sample sets and the results for all 31 proteins the average for age, gender, and education) was used are presented in this manuscript together. The rela- by all cohorts to define cognitively healthy individu- tionship of each individual protein with group-wise als. Further details describing the diagnostic criteria outcome variables was tested using logistic regres- used by each cohort can be found in Bos et al. [17]. sion. The relationship of proteins with continuous For those individuals with MCI at baseline, we outcome variables was examined using linear regres- identified their clinical diagnosis at the last avail- sion. able follow-up visit provided by each local site ROC analysis was performed on each protein (average = 2.21 years). Those that changed in their individually, and the outcome was the dichotomous diagnosis to AD-type dementia, or had a decrease of amyloid status. Sensitivity, specificity and area under at least 3 Mini-Mental State Examination (MMSE) the curve (AUC) metrics were computed. The 95% points, were classed as MCI-convertors. MMSE was confidence intervals were estimated using the boot- provided for all individuals. strap resampling method with n = 1000 repetitions [33]. The optimal operation point from the ROC curve Apolipoprotein E (APOE) genotyping was identified by maximizing the Youden’s J statistic [34]. APOE genotypes for single nucleotide polymor- phisms (SNPs) rs429358 and rs7412 were either Multivariate amyloid classifier determined from genome-wide SNP genotyping data Logistic regression was used to assess the perfor- (generated via the Infinium Global Screening Array mance of a multi-protein model for the discrimination (GSA) with Shared Custom Content, Illumina Inc.) between individuals in the abnormal and normal brain or by TaqMan assays (ThermoFisher Scientific, Fos- amyloid groups. The AUC, sensitivity, specificity, ter City, CA) on a QuantStudio-12K-Flex system and likelihood ratio (LR) of the model are reported. in 384-well format. After QC, there were sufficient The optimal operation point from the ROC curve was data to classify 926 individuals as APOE ␧4 carriers identified by maximizing the Youden’s J statistic [34]. 218 S. Westwood et al. / Validation of Plasma Biomarkers of Aβ Table 1 Demographics of subjects from the EMIF-AD MBD study Variable Subjects included in Subjects included in Oxford sample set Cardiff sample set Normal Abnormal p Low High p amyloid amyloid amyloid amyloid status status status status N 457 543 / 460 406 / A␤ Z-score 0.49 ± 0.62 –1.35 ± 0.48 <0.001∗ 0.45 ± 0.62 –1.36 ± 0.51 <0.001∗ Age (y) 66.52 ± 8.71 69.81 ± 8.12 <0.001∗ 66.61 ± 8.25 70.47 ± 8.22 <0.001∗ Female gender N (%) 223 (49) 301 (55) <0.05∗ 253 (55) 226 (56) 0.844 CTL N (%) 289 (63) 119 (22) / 272 (59) 88 (22) / MCI N (%) 147 (32) 253 (47) / 179 (39) 234 (58) / AD N (%) 21 (5) 171 (31) / 9 (2) 84 (21) / APOE genotype ␧4+ N (%) 147 (32) 349 (64) <0.001∗ 134 (29) 246 (61) <0.001∗ MMSE 27.93 ± 2.49 25.11 ± 4.29 <0.001∗ 28.09 ± 2.23 25.41 ± 4.26 <0.001∗ MCI, mild cognitive impairment; AD, Alzheimer’s disease, APOE, apolipoprotein E; CTL, cognitively healthy control; MMSE, Mini-Mental State Examination. Mean ± standard deviation. ∗ Statistically significant by Mann-Whitney U, p < 0.05. RESULTS in vivo AD pathology [35]. CFI was previously found as a biomarker for conversion from MCI to dementia, Clinical characteristics and inter-group with decreased plasma protein concentration in MCI differences converters [36]. Therefore, the direction of change identified in this study, decreased CFI with increased The clinical characteristics of the Oxford and pathology, agrees with this previous finding. Cardiff sample sets, stratified by amyloid status, Logistic regression analysis was also performed are presented in Table 1. Across both sample separately for each diagnostic group and APOE sets, individuals with abnormal amyloid status were ␧4 carrier groups (␧4 non-carrier/␧4 carrier). These older (p < 0.001), more frequently APOE ␧4 carriers results are reported in Supplementary Tables 1–5. (p < 0.001), and had lower MMSE scores (p < 0.001) ROC analysis was performed on each of the seven compared to those with normal amyloid status. proteins found to replicate direction of change to Within the Oxford set only, individuals with abnor- determine their individual predictive ability for the mal amyloid status were more frequently female discrimination of normal/abnormal amyloid status (p < 0.05) compared to those with normal amyloid groups, and compared to the discriminant ability of status. age. Table 3 displays the results of the ROC analysis and Supplementary Figure 2 displays both the AUC Univariate statistics for amyloid status (high/low and corresponding 95% confidence interval of each group) protein and age. The AUC for every protein is higher than chance, even when including the lower end of Cross-sectional comparisons of protein concen- the confidence interval. trations between normal and abnormal amyloid For these seven proteins, ROC analysis of amyloid status groups revealed seven proteins that remained status classification was also performed separately for statistically significant after multiple testing correc- each diagnostic group and APOE ␧4 carrier groups tions (q < 0.05; ficolin-2 (FCN2), ␤2-microglobulin (␧4 non-carrier/␧4 carrier). These results are reported (B2M), alpha-1 antitrypsin (A1AT), apolipoprotein in Supplementary Tables 6 and 7. E (apoE), complement component 4 (C4), Cathepsin D (CTSD), complement factor I (CFI), Table 2). Five of these proteins have previously been discovered as Multi-protein classifier of amyloid biomarkers of in vivo brain amyloid pathology in one normal/abnormal status or more of our previous biomarker studies (FCN2, B2M, A1AT, apoE, and C4), and all five replicate It is possible, if the individual protein associa- the direction of change previously identified. Expres- tions with in vivo AD pathology are independent sion of cathepsin D has previously been found to be of each other, that a compound marker set of some decreased in AD fibroblasts, and here we show that or all of these proteins would have greater predic- this protein is also decreased in plasma with increased tive value than any one protein alone. In order to S. Westwood et al. / Validation of Plasma Biomarkers of Aβ 219 Table 2 Logistic regression (age as covariate) results for each protein with amyloid status as the outcome variable Sample protein Logistic Regression set beta p q N FCN2 0.466 0.000∗ 0.000∗ 824 FGG –0.071 0.300 0.521 891 Cystatin C –0.135 0.049∗ 0.124 898 Clusterin –0.138 0.045∗ 0.124 906 B2M –0.266 0.000∗ 0.004∗ 832 AGP 0.009 0.901 0.963 875 CP 0.028 0.679 0.780 883 A2M –0.017 0.813 0.900 896 ApoA1 –0.073 0.290 0.521 886 ApoC3 –0.001 0.991 0.991 900 apoE –0.220 0.002∗ 0.008∗ 901 Oxford TTR –0.039 0.572 0.772 901 CFH –0.049 0.475 0.701 907 CRP –0.139 0.049∗ 0.124 854 A1AT –0.213 0.004∗ 0.016∗ 784 PEDF –0.006 0.933 0.964 853 SAP –0.052 0.449 0.697 869 CC4 0.243 0.001∗ 0.005∗ 894 BDNF –0.036 0.603 0.778 887 Cathepsin D –0.238 0.001∗ 0.005∗ 880 sICAM-1 –0.148 0.033∗ 0.124 893 RANTES –0.100 0.146 0.309 905 NCAM 0.032 0.645 0.780 891 sVCAM-1 –0.071 0.302 0.521 891 PAI.1 –0.134 0.052 0.124 890 CR1 –0.077 0.378 0.617 770 TCC 0.112 0.149 0.309 770 Cardiff CFB –0.042 0.567 0.772 788 CFI –0.284 0.000∗ 0.004∗ 754 Eotaxin 0.176 0.037∗ 0.124 749 MCP 0.031 0.680 0.780 769 ∗ Statistically significant < 0.05. Table 3 AUC statistics per protein, for the classification of normal / abnormal brain amyloid status. Abbreviations: AUC, area under the curve; CI, 95% confidence interval Variable Optimal cutpoint Sensitivity Specificity AUC CI. CI. low high FCN2 24607094.990 0.448 0.783 0.640 0.593 0.690 Age 67.355 0.641 0.560 0.619 0.578 0.659 CFI 26793.107 0.547 0.623 0.585 0.531 0.639 C4 73789.784 0.436 0.713 0.580 0.533 0.630 B2M 4232.708 0.433 0.698 0.577 0.526 0.628 Cathepsin D 322.630 0.586 0.547 0.576 0.528 0.625 apoE 106.020 0.667 0.445 0.554 0.504 0.602 A1AT 1675622.528 0.765 0.333 0.552 0.506 0.606 test this, the seven proteins significant by logistic tested on 374 individuals and achieved moderate regression after multiple testing corrections (q < 0.05; accuracy (AUC = 0.742 (Fig. 1, sensitivity = 0.682, FCN2, B2M, apoE, A1AT, CC4, cathepsin D, CFI) specificity = 0.704, LR = 2.3). In comparison, age were included in a logistic regression classifier, alone achieved an AUC = 0.617. along with age, to determine their predictive ability To determine how this 8-feature model performs for amyloid status when combined. After miss- at different stages of the disease process and also in ing data were removed this 8-feature model was APOE ␧4 carriers and non-carriers independently, we 220 S. Westwood et al. / Validation of Plasma Biomarkers of Aβ Relationship of classifier proteins with other markers of in vivo AD pathology or disease progression In order to determine whether the classifier pro- teins were specific to brain amyloid pathology or if they could also perform as a biomarker of the other key hallmark of AD, brain tau pathology, linear regression was used to assess the continuous rela- tionship with both p-tau and t-tau Z-scores (Table 5). None of the proteins were significantly related to either measure after multiple testing correction. We then used a similar approach to examine their rela- tionship to hippocampal volume and MMSE score, and used logistic regression to examine their predic- tive ability for MCI conversion to dementia (Table 5). FCN2 displayed a significant relationship with all Fig. 1. ROC curve obtained for the 8-feature classifier for predic- three measures (q < 0.05). CFI was found to be sig- tion of amyloid normal/abnormal status. The 8-features included were: FCN2, B2M, apoE, A1AT, CC4, cathepsin D, CFI, and age. nificantly related to both baseline MMSE and MCI conversion to dementia (q < 0.05). A1AT was found to be significantly related to hippocampal volume only (q < 0.05). tested the classification ability of this model within each separate diagnostic group and APOE ␧4 carriers and non-carriers. Table 4 displays the performance DISCUSSION of this model within each group. The performance within the AD-type dementia group could not be We have previously used a pathology endophe- accurately determined since removal of subjects with notype approach to discover plasma proteomic missing data left only 5% of the AD-type dementia biomarkers designed to be predictive of in vivo AD cases as amyloid negative. pathology and disease progression. The aim of the present study was to replicate these previously iden- tified candidate biomarkers in a large multi-center Relationship of classifier proteins with pragmatic sample collection collated from multiple continuous Aβ Z-score cohorts, as well as to identify a plasma proteomic panel that could classify individuals into high or low We next wanted to determine whether this protein- brain amyloid groups. Our results show that in around amyloid relationship remains consistent using the A␤ 1,000 samples from multiple studies across Europe, Z-score (Table 5). All, except one protein (A1AT), seven biomarkers replicate, and we confirm a panel were significantly related to A␤ Z-score after testing of proteins that identify abnormal levels of in vivo for multiple testing corrections (q < 0.05). A1AT was amyloid pathology with a sensitivity of 0.682 and a tending towards significance (q = 0.061). specificity of 0.704. While we chose to maximize the Table 4 ROC and AUC statistics for the 8-feature model for the classification of amyloid status within each diagnostic group, and APOE ␧4 carriers and non-carriers AUC Sensitivity Specificity LR N Whole cohort (AD, MCI, CTL) 0.742 0.682 0.704 2.30 374 MCI 0.743 0.658 0.785 3.06 193 CTL 0.768 0.682 0.776 3.04 142 MCI and CTL combined 0.724 0.677 0.661 2.00 335 APOE ␧4 non-carriers 0.736 0.681 0.725 2.47 199 APOE ␧4 carriers 0.836 0.757 0.767 3.25 175 MCI, mild cognitive impairment; AD, Alzheimer’s disease, APOE, apolipoprotein E; CTL, cogni- tively healthy control; AUC, area under the curve; LR, likelihood ratio. S. Westwood et al. / Validation of Plasma Biomarkers of Aβ 221 Youden’s index, which gave a balanced compromise 0.006∗ 0.000∗ 0.901 0.901 0.901 0.483 0.901 between false negative and false positive results, simi- q Logistic regression MCI conversion to larly, any other operational point may also be suitable, dementia depending on the clinical utility and priorities. More- 0.002∗ –0.932 0.000∗ 0.729 0.901 0.207 0.792 0.52 p over, the model has predictive value in both APOE 0.398 ␧4 carrier and non-carrier individuals generating a 0.039 –0.015 –0.096 0.153 0.032 biological predictor that could be used to reduce the ␤ screen failure rate in clinical trials. 0.000∗ 0.002∗ The final 7-protein biomarker panel consists 0.214 0.204 0.181 0.306 0.204 q of: ␤2-microglobulin (B2M), cathepsin D (CTSD), Linear regression ficolin-2 (FCN2), complement component 4 (C4), MMSE 0.000∗ 0.442 0.001∗ 0.183 0.146 0.078 0.306 0.118 alpha-1 antitrypsin (A1AT), complement factor I p Linear and logistic regression (age as covariate) results for each protein with alternative outcome measures (CFI), and apolipoprotein E (apoE). We did not find 0.174 –0.184 0.001∗ 0.239 0.261 –0.131 0.195 any significant improvement to AUC by introducing 0.000∗ –0.88 ␤ standardized quadratic terms for each of the signifi- cant variables. Specifically, our variables are: ‘Age’ 0.758 0.429 0.429 0.47 q and seven proteins: ‘FCN2’, ‘B2M’, ‘ApoE’, ‘A1AT’, Linear regression ‘CC4’, CathepsinD’, and ‘CFI’. Using them as a lin- Hippocampal 0.000∗ 0.000∗ 0.758 0.403 0.112 0.306 52.396 0.256 ear combination, we achieved AUC = 0.742 and the volume p most significant ones (p < 0.001) were Age, FCN2, 37.388 172.168 –70.691 46.917 and B2M. Their quadratic terms were found to be –244.14 –14.6 both, small (effect size) and statistically insignifi- ␤ cant. Although initial discovery-phase studies used an unbiased approach to identify proteins that have 0.132 0.991 0.778 0.991 0.778 0.991 0.221 a relationship with AD and its pathology, it is note- q Linear regression Table 5 T-Tau Z-score worthy that the resulting biomarker candidates are also biologically relevant to the disease process. B2M 0.019∗ 0.991 0.444 0.924 0.445 0.776 0.107 0.063 p shares structural characteristics with fibrillar A␤ [37] and is one of a number of proteins that form amyloid –0.138 –0.001 0.045 –0.006 –0.045 0.017 deposits. Cathepsin D is increased in tangle bearing ␤ neurons [38], a process that might be induced by A␤ [39]. Ficolin-2 and mannose-binding lectin (MBL) 0.098 0.876 0.584 0.876 0.209 0.876 0.098 q Linear regression are both activators of the lectin complement pathway P-Tau Z-score [40] and CSF MBL levels have been shown to be 0.017∗ 0.116 0.028∗ 0.715 0.334 0.843 0.876 0.09 reduced in AD [41]. The complement proteins (C4 p and CFI) are two members of a pathway repeatedly –0.125 0.019 –0.011 –0.088 –0.008 shown to be associated with AD through genetics and 0.05 ␤ neuropathology as well as from biomarker studies [42]. A1AT is an acute phase chemoreactant that is 0.000∗ 0.001∗ 0.001∗ 0.001∗ 0.016∗ 0.117 0.001∗ 0.002∗ 0.061 q metabolized by the serpin enzyme complex (SEC) Linear regression and hence might compete with and affect the activ- A␤ Z-score 0.000∗ 0.001∗ 0.000∗ 0.001∗ 0.013∗ 0.061 ity of another SEC ligand, A␤ [43]. The relationship p between the APOE ␧2/␧3/␧4 polymorphism and AD 0.125 0.123 –0.117 0.086 is well established and complementary to our finding –0.19 0.07 significant < 0.05. ␤ here that apoE protein is clearly a marker of amyloid load. Cathepsin D Recently, Nakamura et al. (2018) published an important study demonstrating that A␤ fragments Sample protein FCN2 A1AT B2M apoE CC4 CFI can be reliably detected in blood and perform as ∗ Statistically well as current CSF biomarkers for predicting brain amyloid [44]. This approach using immunoprecipi- set Oxford Cardiff tation combined with mass spectrometry (IP+MS), 222 S. Westwood et al. / Validation of Plasma Biomarkers of Aβ was able to predict brain amyloid status with up to amyloid positivity rate of 30% and a representative 90% accuracy. This finding builds upon other recent cost of $3500 for an amyloid PET scan, recruiting work by Ovod et al. (2017) who also used an IP+MS 1,000 amyloid positive subjects would require 3,333 technique to identify blood A␤ with high concor- subjects at a total cost of $11.7 million. Pre-screening dance to amyloid PET [45]. While these papers are can reduce this cost. Utilizing the optimal model sen- important in demonstrating the value of blood A␤ sitivity and specificity values (Table 4) and assuming as a biomarker tool for brain pathology, replication an assay cost of $150, the total cost of recruiting in more studies with larger sample sizes is needed, would be $7.8 million and require pre-screening 4900 and the proteomic technology employed will require subjects. Going further, model sensitivities and speci- refinement to enable implementation at scale. Nev- ficities are based on cut-offs that can be optimized ertheless, these studies provide considerable further to be fit-for-purpose for a given application. For the proof of concept for AD blood biomarkers. While the present model, optimizing the cut-off for minimum accuracy rates reported in our study are not as high as total cost would yield a sensitivity of 34% and a speci- those reported when measuring blood A␤ levels, we ficity of 93% requiring pre-screening 9700 subjects are still able to achieve a level of accuracy sufficient with a total recruitment cost of $6.6 million. When to make a substantial impact upon the cost efficiency the positive prevalence is low and the cost of the truth of clinical trials while using immunocapture—a prac- assay is high relative to the pre-screen, such economic tical and low-cost assay technology already in very optimizations often favor specificity to reject a large wide use in both clinical and research laboratories. portion of true negatives to save operating costs. Such Given that sample collection and storage pro- economic trade-offs require a larger screening popu- tocols differed across the 11 cohorts included in lation and have a higher false negative rate. Screening the EMIF-AD MBD study [17], this pragmatic in medical practice, however, favors higher sensitivity meta-collection reflects the challenges faced by any to optimize the yield of subjects for intervention. In putative biomarker in the real-world of multi-site, short, low-cost pre-screening applications can save multi-national clinical studies, and even more so in money and be tuned as needed for the application clinical practice, where standardization of sample context. collection is sought but rarely achieved. Replication To summarize, our overall goal is to facilitate of any putative biomarker set in such a collection of clinical trials by contributing to rapid and effective samples has a higher prior probability of effective selection of research participants most likely to have utility in practice than replication in a single cohort brain amyloid pathology and hence reducing screen or single site study with a fully standardized sam- failure rates, reducing cost and time of trial start- ple collection procedure. Nonetheless, the limitations up and reducing exposure of potential participants to of this study are also acknowledged; this study was PET imaging or CSF lumbar puncture. In order to do designed to determine whether candidate biomark- this, we have previously investigated, discovered, and ers replicate in their ability to predict amyloid, it replicated plasma protein biomarkers that could be was not designed to determine the real-world value implemented as a clinical trial entry criterion to triage of the biomarkers or the biomarker panel they form potential participants for amyloid PET or CSF mea- for clinical trial screening. Selecting half of the sam- sures. In the current study, seven of these biomarkers ples from people with high amyloid enables proof of are replicated in a large, multi-center cohort. These concept but clearly random prediction will already seven proteins form a biomarker panel that is the identify 50% of those with amyloid correctly. Addi- product of over a decade of research, is biologically tionally, missing data (4–11% per protein as a result relevant and measurable using practical immunocap- of variable sample volume and quality and occasional ture arrays, and could significantly reduce the cost assay performance failure) meant that when combin- incurred to clinical trials by screen failure because of ing multiple proteins together in a biomarker panel absence of amyloid pathology. the overall sample size with complete data was sig- nificantly reduced. However, even in this reduced sample set with 50% prediction accuracy possible by ACKNOWLEDGMENTS chance, the likelihood ratio with our biomarker panel is 2.3 in the whole cohort and 3 for preclinical disease, This research was conducted as part of the EMIF- suggesting approximately a 15% and 20% improve- AD project which has received support from the ment in detection respectively [46]. With a typical Innovative Medicines Initiative Joint Undertaking S. Westwood et al. / Validation of Plasma Biomarkers of Aβ 223 under EMIF grant agreement n◦ 115372, resources [2] Cummings J, Lee G, Ritter A, Sabbagh M, Zhong K of which are composed of financial contribution (2019) Alzheimer’s disease drug development pipeline: 2019. Alzheimers Dement (N Y) 5, 272-293. from the European Union’s Seventh Framework Pro- [3] Jansen WJ, Ossenkoppele R, Knol DL, Tijms BM, Scheltens gramme (FP7/2007–2013) and EFPIA companies’ P, Verhey FR, Visser PJ, Aalten P, Aarsland D, Alcolea D, in-kind contribution. ANH and SL were supported Alexander M, Almdahl IS, Arnold SE, Baldeiras I, Barthel by a MRC Mental Health Data Pathfinder award to H, van Berckel BN, Bibeau K, Blennow K, Brooks DJ, van Buchem MA, Camus V, Cavedo E, Chen K, Chete- the University of Oxford, and by the NIHR Oxford lat G, Cohen AD, Drzezga A, Engelborghs S, Fagan AM, Health Biomedical Research Centre (BRC). FB is Fladby T, Fleisher AS, van der Flier WM, Ford L, Forster supported by the NIHR UCLH biomedical research S, Fortea J, Foskett N, Frederiksen KS, Freund-Levi Y, centre. The Lausanne study was funded by a grant Frisoni GB, Froelich L, Gabryelewicz T, Gill KD, Gkatzima O, Gomez-Tortosa E, Gordon MF, Grimmer T, Hampel H, from the Swiss National Research Foundation (SNF Hausner L, Hellwig S, Herukka SK, Hildebrandt H, Ishi- 320030 141179). Cardiff University acknowledge hara L, Ivanoiu A, Jagust WJ, Johannsen P, Kandimalla R, Wellcome Trust support (grant: 104025/Z/14/Z) as Kapaki E, Klimkowicz-Mrowiec A, Klunk WE, Kohler S, part of the research project entitled “Wellcome Trust Koglin N, Kornhuber J, Kramberger MG, Van Laere K, Lan- dau SM, Lee DY, de Leon M, Lisetti V, Lleo A, Madsen K, Consortium for Neuroimmunology of Mood Disor- Maier W, Marcusson J, Mattsson N, de Mendonca A, Meu- ders and Alzheimer’s Disease”. The views expressed lenbroek O, Meyer PT, Mintun MA, Mok V, Molinuevo JL, are those of the authors and not necessarily those of Mollergard HM, Morris JC, Mroczko B, Van der Mussele S, Na DL, Newberg A, Nordberg A, Nordlund A, Novak GP, the MRC, the NHS, the NIHR or the Department of Paraskevas GP, Parnetti L, Perera G, Peters O, Popp J, Prab- Health. hakar S, Rabinovici GD, Ramakers IH, Rami L, Resende de Authors’ disclosures available online (https:// Oliveira C, Rinne JO, Rodrigue KM, Rodriguez-Rodriguez www.j-alz.com/manuscript-disclosures/19-0434r3). E, Roe CM, Rot U, Rowe CC, Ruther E, Sabri O, Sanchez- Juan P, Santana I, Sarazin M, Schroder J, Schutte C, Seo SW, Soetewey F, Soininen H, Spiru L, Struyfs H, Teunissen CE, ETHICS STATEMENT Tsolaki M, Vandenberghe R, Verbeek MM, Villemagne VL, Vos SJ, van Waalwijk van Doorn LJ, Waldemar G, Wallin A, Wallin AK, Wiltfang J, Wolk DA, Zboch M, Zetter- Written informed consent was obtained from all berg H (2015) Prevalence of cerebral amyloid pathology participants before inclusion in the study. The exper- in persons without dementia: A meta-analysis. JAMA 313, iments were done in accordance with the Helsinki 1924-1938. [4] Hye A, Lynham S, Thambisetty M, Causevic M, Campbell Declaration of 1975 and the medical ethics commit- J, Byers HL, Hooper C, Rijsdijk F, Tabrizi SJ, Banner S, tee at each site approved the study (Supplementary Shaw CE, Foy C, Poppe M, Archer N, Hamilton G, Powell J, Table 8). Brown RG, Sham P, Ward M, Lovestone S (2006) Proteome- based plasma biomarkers for Alzheimer’s disease. Brain 129, 3042-3050. DATA AVAILABILITY [5] Humpel C (2011) Identifying and validating biomarkers for Alzheimer’s disease. Trends Biotechnol 29, 26-32. [6] Henriksen K, O’Bryant SE, Hampel H, Trojanowski JQ, The datasets generated and analyzed during the Montine TJ, Jeromin A, Blennow K, Lonneborg A, Wyss- current study are available via the EMIF-AD Cata- Coray T, Soares H, Bazenet C, Sjogren M, Hu W, Lovestone logue via submitted research questions which have S, Karsdal MA, Weiner MW (2014) The future of blood- to be approved by the data-owners from each parent based biomarkers for Alzheimer’s disease. Alzheimers Dement 10, 115-131. cohort. [7] O’Bryant SE, Gupta V, Henriksen K, Edwards M, Jeromin A, Lista S, Bazenet C, Soares H, Lovestone S, Hampel H, Montine T, Blennow K, Foroud T, Carrillo M, Graff- SUPPLEMENTARY MATERIAL Radford N, Laske C, Breteler M, Shaw L, Trojanowski JQ, Schupf N, Rissman RA, Fagan AM, Oberoi P, Umek The supplementary material is available in the elec- R, Weiner MW, Grammas P, Posner H, Martins R (2015) tronic version of this article: https://dx.doi.org/10. Guidelines for the standardization of preanalytic variables for blood-based biomarker studies in Alzheimer’s disease 3233/JAD-190434. research. Alzheimers Dement 11, 549-560. [8] Thambisetty M, Simmons A, Velayudhan L, Hye A, Camp- bell J, Zhang Y, Wahlund LO, Westman E, Kinsey A, REFERENCES Guntert A, Proitsi P, Powell J, Causevic M, Killick R, Lun- non K, Lynham S, Broadstock M, Choudhry F, Howlett [1] Cummings J, Lee G, Ritter A, Zhong K (2018) Alzheimer’s DR, Williams RJ, Sharp SI, Mitchelmore C, Tunnard C, disease drug development pipeline: 2018. Alzheimers Leung R, Foy C, O’Brien D, Breen G, Furney SJ, Ward M, Dement (N Y) 4, 195-214. Kloszewska I, Mecocci P, Soininen H, Tsolaki M, Vellas B, 224 S. Westwood et al. / Validation of Plasma Biomarkers of Aβ Hodges A, Murphy DG, Parkins S, Richardson JC, Resnick Inflammatory biomarkers in Alzheimer’s disease plasma. SM, Ferrucci L, Wong DF, Zhou Y, Muehlboeck S, Evans Alzheimers Dement 15, 776-787. A, Francis PT, Spenger C, Lovestone S (2010) Association [17] Bos I, Vos S, Vandenberghe R, Scheltens P, Engelborghs of plasma clusterin concentration with severity, pathology, S, Frisoni G, Molinuevo JL, Wallin A, Lleo A, Popp J, and progression in Alzheimer disease. Arch Gen Psychiatry Martinez-Lage P, Baird A, Dobson R, Legido-Quigley C, 67, 739-748. Sleegers K, Van Broeckhoven C, Bertram L, Ten Kate M, [9] Kiddle SJ, Thambisetty M, Simmons A, Riddoch-Contreras Barkhof F, Zetterberg H, Lovestone S, Streffer J, Visser J, Hye A, Westman E, Pike I, Ward M, Johnston C, Lup- PJ (2018) The EMIF-AD Multimodal Biomarker Discov- ton MK, Lunnon K, Soininen H, Kloszewska I, Tsolaki M, ery study: Design, methods and cohort characteristics. Vellas B, Mecocci P, Lovestone S, Newhouse S, Dobson R, Alzheimers Res Ther 10, 64. Alzheimers Disease Neuroimaging Initiative (2012) Plasma [18] Visser PJ, Verhey FRJ, Boada M, Bullock R, De Deyn PP, based markers of [11C] PiB-PET brain amyloid burden. Frisoni GB, Frolich L, Hampel H, Jolles J, Jones R, Minthon PLoS One 7, e44260. L, Nobili F, Rikkert MO, Ousset PJ, Rigaud AS, Scheltens [10] Hye A, Riddoch-Contreras J, Baird AL, Ashton NJ, Bazenet P, Soininen H, Spiru L, Touchon J, Tsolaki M, Vellas B, C, Leung R, Westman E, Simmons A, Dobson R, Sattlecker Wahlund LO, Wilcock G, Winblad B (2008) Development M, Lupton M, Lunnon K, Keohane A, Ward M, Pike I, Zucht of screening guidelines and clinical criteria for predementia HD, Pepin D, Zheng W, Tunnicliffe A, Richardson J, Gau- Alzheimer’s disease. Neuroepidemiology 30, 254-265. thier S, Soininen H, Kloszewska I, Mecocci P, Tsolaki M, [19] Reijs BLR, Ramakers IHGB, Elias-Sonnenschein L, Teu- Vellas B, Lovestone S (2014) Plasma proteins predict con- nissen CE, Koel-Simmelink M, Tsolaki M, Wahlund LO, version to dementia from prodromal disease. Alzheimers Waldemar G, Hausner L, Johannsen P, Vanderstichele H, Dement 10, 799-807.e792. Verhey F, Devanand DP, Visser PJ (2017) Relation of [11] Sattlecker M, Kiddle SJ, Newhouse S, Proitsi P, Nelson odor identification with Alzheimer’s disease markers in S, Williams S, Johnston C, Killick R, Simmons A, West- cerebrospinal fluid and cognition. J Alzheimers Dis 60, man E, Hodges A, Soininen H, Kloszewska I, Mecocci 1025-1034. P, Tsolaki M, Vellas B, Lovestone S, AddNeuroMed Con- [20] Galluzzi S, Marizzoni M, Babiloni C, Albani D, Antelmi sortium, Dobson RJ (2014) Alzheimer’s disease biomarker L, Bagnoli C, Bartres-Faz D, Cordone S, Didic M, Farotti discovery using SOMAscan multiplexed protein technol- L, Fiedler U, Forloni G, Girtler N, Hensch T, Jovicich J, ogy. Alzheimers Dement 10, 724-734. Leeuwis A, Marra C, Molinuevo JL, Nobili F, Pariente J, [12] Ashton NJ, Kiddle SJ, Graf J, Ward M, Baird AL, Hye A, Parnetti L, Payoux P, Del Percio C, Ranjeva JP, Rolandi E, Westwood S, Wong KV, Dobson RJ, Rabinovici GD, Miller Rossini PM, Schonknecht P, Soricelli A, Tsolaki M, Visser BL, Rosen HJ, Torres A, Zhang Z, Thurfjell L, Covin A, PJ, Wiltfang J, Richardson JC, Bordet R, Blin O, Frisoni Hehir CT, Baker D, Bazenet C, Lovestone S, AIBL Research GB, PharmaCog Consortium (2016) Clinical and biomarker Group (2015) Blood protein predictors of brain amyloid for profiling of prodromal Alzheimer’s disease in workpackage enrichment in clinical trials? Alzheimers Dement (Amst) 1, 5 of the Innovative Medicines Initiative PharmaCog project: 48-60. A ‘European ADNI study’. J Intern Med 279, 576-591. [13] Voyle N, Baker D, Burnham SC, Covin A, Zhang Z, Sangur- [21] van der Flier WM, Pijnenburg YAL, Prins N, Lemstra AW, dekar DP, Tan Hehir CA, Bazenet C, Lovestone S, Kiddle Bouwman FH, Teunissen CE, van Berckel BNM, Stam S, Dobson RJ, AIBL research group (2015) Blood protein CJ, Barkhof F, Visser PJ, van Egmond E, Scheltens P markers of neocortical amyloid-beta burden: A candidate (2014) Optimizing patient care and research: The Amster- study using SOMAscan technology. J Alzheimers Dis 46, dam Dementia Cohort. J Alzheimers Dis 41, 313-327. 947-961. [22] Somers C, Struyfs H, Goossens J, Niemantsverdriet E, Luy- [14] Westwood S, Leoni E, Hye A, Lynham S, Khondoker MR, ckx J, De Roeck N, De Roeck E, De Vil B, Cras P, Martin Ashton NJ, Kiddle SJ, Baird AL, Sainz-Fuertes R, Leung R, JJ, De Deyn PP, Bjerke M, Engelborghs S (2016) A decade Graf J, Hehir CT, Baker D, Cereda C, Bazenet C, Ward M, of cerebrospinal fluid biomarkers for Alzheimer’s disease Thambisetty M, Lovestone S (2016) Blood-based biomarker in Belgium. J Alzheimers Dis 54, 383-395. candidates of cerebral amyloid using PiB PET in non- [23] Estanga A, Ecay-Torres M, Ibanez A, Izagirre A, Villanua demented elderly. J Alzheimers Dis 52, 561-572. J, Garcia-Sebastian M, Gaspar MTI, Otaegui-Arrazola A, [15] Westwood S, Baird AL, Hye A, Ashton NJ, Nevado- Iriondo A, Clerigue M, Martinez-Lage P (2017) Beneficial Holgado AJ, Anand SN, Liu B, Newby D, Bazenet C, Kiddle effect of bilingualism on Alzheimer’s disease CSF biomark- SJ, Ward M, Newton B, Desai K, Tan Hehir C, Zanette M, ers and cognition. Neurobiol Aging 50, 144-151. Galimberti D, Parnetti L, Lleo A, Baker S, Narayan VA, [24] Wallin A, Nordlund A, Jonsson M, Blennow K, Zetterberg van der Flier WM, Scheltens P, Teunissen CE, Visser PJ, H, Ohrfelt A, Stalhammar J, Eckerstrom M, Carlsson M, Lovestone S (2018) Plasma protein biomarkers for the pre- Olsson E, Gothlin M, Svensson J, Rolstad S, Eckerstrom C, diction of CSF amyloid and tau and [(18)F]-Flutemetamol Bjerke M (2016) Alzheimer’s diseasesubcortical vascular PET scan result. Front Aging Neurosci 10, 409. disease spectrum in a hospital-based setting: Overview of [16] Morgan AR, Touchard S, Leckey C, O’Hagan C, Nevado- results from the Gothenburg MCI and dementia studies. J Holgado AJ, Barkhof F, Bertram L, Blin O, Bos I, Dobricic Cereb Blood Flow Metab 36, 95-113. V, Engelborghs S, Frisoni G, Frolich L, Gabel S, Johannsen [25] Fortea J, Sala-Llonch R, Bartres-Faz D, Bosch B, Llado P, Kettunen P, Kloszewska I, Legido-Quigley C, Lleo A, A, Bargallo N, Molinuevo JL, Sanchez-Valle R (2010) Martinez-Lage P, Mecocci P, Meersmans K, Molinuevo JL, Increased cortical thickness and caudate volume precede Peyratout G, Popp J, Richardson J, Sala I, Scheltens P, Stref- atrophy in PSEN1 mutation carriers. J Alzheimers Dis 22, fer J, Soininen H, Tainta-Cuezva M, Teunissen C, Tsolaki M, 909-922. Vandenberghe R, Visser PJ, Vos S, Wahlund LO, Wallin A, [26] Tautvydaite D, Kukreja D, Antonietti JP, Henry H, von Gun- Westwood S, Zetterberg H, Lovestone S, Morgan BP (2019) ten A, Popp J (2017) Interaction between personality traits S. Westwood et al. / Validation of Plasma Biomarkers of Aβ 225 and cerebrospinal fluid biomarkers of Alzheimer’s disease [36] Hakobyan S, Harding K, Aiyaz M, Hye A, Dobson R, Baird pathology modulates cognitive performance. Alzheimers A, Liu B, Harris CL, Lovestone S, Morgan BP (2016) Com- Res Ther 9, 6. plement biomarkers as predictors of disease progression in [27] Adamczuk K, Schaeverbeke J, Nelissen N, Neyens V, Van- Alzheimer’s disease. J Alzheimers Dis 54, 707-716. denbulcke M, Goffin K, Lilja J, Hilven K, Dupont P, Van [37] Sawaya MR, Sambashivan S, Nelson R, Ivanova MI, Sievers Laere K, Vandenberghe R (2016) Amyloid imaging in SA, Apostol MI, Thompson MJ, Balbirnie M, Wiltzius JJ, cognitively normal older adults: Comparison between F- McFarlane HT, Madsen AO, Riekel C, Eisenberg D (2007) 18-flutemetamol and C-11-Pittsburgh compound B. Eur J Atomic structures of amyloid cross-beta spines reveal varied Nucl Med Mol Imaging 43, 142-151. steric zippers. Nature 447, 453-457. [28] Alcolea D, Carmona-Iragui M, Suarez-Calvet M, Sanchez- [38] Ginsberg SD, Hemby SE, Lee VM, Eberwine JH, Tro- Saudinos MB, Sala I, Anton-Aguirre S, Blesa R, Clarimon janowski JQ (2000) Expression profile of transcripts in J, Fortea J, Lleo A (2014) Relationship between beta- Alzheimer’s disease tangle-bearing CA1 neurons. Ann Neu- secretase, inflammation and core cerebrospinal fluid rol 48, 77-87. biomarkers for Alzheimer’s disease. J Alzheimers Dis 42, [39] Hoffman KB, Bi XN, Pham JT, Lynch G (1998) ␤-amyloid 157-167. increases cathepsin D levels in hippocampus. Neurosci Lett [29] Scheufele E, Aronzon D, Coopersmith R, McDuffie MT, 250, 75-78. Kapoor M, Uhrich CA, Avitabile JE, Liu J, Housman D, [40] Fujita T, Matsushita M, Endo Y (2004) The lectin- Palchuk MB (2014) tranSMART: An open source knowl- complement pathway - its role in innate immunity and edge management and high content data analytics platform. evolution. Immunol Rev 198, 185-202. AMIA Jt Summits Transl Sci Proc 2014, 96-101. [41] Lanzrein AS, Jobst KA, Thiel S, Jensenius JC, Sim RB, [30] McKhann G, Drachman D, Folstein M, Katzman R, Price D, Perry VH, Sim E (1998) Mannan-binding lectin in human Stadlan EM (1984) Clinical diagnosis of Alzheimer’s dis- serum, cerebrospinal fluid and brain tissue and its role in ease: Report of the NINCDS-ADRDA Work Group under Alzheimer’s disease. Neuroreport 9, 1491-1495. the auspices of Department of Health and Human Services [42] Aiyaz M, Lupton MK, Proitsi P, Powell JF, Lovestone Task Force on Alzheimer’s Disease. Neurology 34, 939-944. S (2012) Complement activation as a biomarker for [31] Petersen RC (2004) Mild cognitive impairment as a diag- Alzheimer’s disease. Immunobiology 217, 204-215. nostic entity. J Intern Med 256, 183-194. [43] Joslin G, Griffin GL, August AM, Adams S, Fallon RJ, [32] Winblad B, Palmer K, Kivipelto M, Jelic V, Fratiglioni L, Senior RM, Perlmutter DH (1992) The serpin-enzyme Wahlund LO, Nordberg A, Backman L, Albert M, Almkvist complex (SEC) receptor mediates the neutrophil chemo- O, Arai H, Basun H, Blennow K, de Leon M, DeCarli C, tactic effect of alpha-1 antitrypsin-elastase complexes and Erkinjuntti T, Giacobini E, Graff C, Hardy J, Jack C, Jorm amyloid-beta peptide. J Clin Invest 90, 1150-1154. A, Ritchie K, van Duijn C, Visser P, Petersen RC (2004) [44] Nakamura A, Kaneko N, Villemagne VL, Kato T, Doecke Mild cognitive impairment–beyond controversies, towards J, Doré V, Fowler C, Li Q-X, Martins R, Rowe C, Tomita a consensus: Report of the International Working Group on T, Matsuzaki K, Ishii K, Ishii K, Arahata Y, Iwamoto S, Mild Cognitive Impairment. J Intern Med 256, 240-246. Ito K, Tanaka K, Masters CL, Yanagisawa K (2018) High [33] DiCiccio T, Tibshirani R (1987) Bootstrap confidence inter- performance plasma amyloid-␤ biomarkers for Alzheimer’s vals and bootstrap approximations. J Am Stat Assoc 82, disease. Nature 554, 249-254. 163-170. [45] Ovod V, Ramsey KN, Mawuenyega KG, Bollinger JG, [34] Ruopp MD, Perkins NJ, Whitcomb BW, Schisterman EF Hicks T, Schneider T, Sullivan M, Paumier K, Holtzman (2008) Youden Index and optimal cut-point estimated from DM, Morris JC, Benzinger T, Fagan AM, Patterson BW, observations affected by a lower limit of detection. Biom J Bateman RJ (2017) Amyloid ␤ concentrations and stable 50, 419-430. isotope labeling kinetics of human plasma specific to cen- [35] Urbanelli L, Emiliani C, Massini C, Persichetti E, Orlacchio tral nervous system amyloidosis. Alzheimers Dement 13, A, Pelicci G, Sorbi S, Hasilik A, Bernardi G, Orlacchio A 841-849. (2008) Cathepsin D expression is decreased in Alzheimer’s [46] McGee S (2002) Simplifying likelihood ratios. J Gen Intern disease fibroblasts. Neurobiol Aging 29, 12-22. Med 17, 646-649.