Alzheimer’s & Dementia 15 (2019) 1478-1488 Featured Article Discovery and validation of plasma proteomic biomarkers relating to brain amyloid burden by SOMAscan assay Liu Shia,*, Sarah Westwooda, Alison L. Bairda, Laura Winchestera, Valerija Dobricicb, Fabian Kilpertb, Shengjun Hongb, Andre Frankec, Abdul Hyed, Nicholas J. Ashtond,e,f,g, Angharad R. Morganh, Isabelle Bosi, Stephanie J. B. Vosj, Noel J. Buckleya, Mara ten Katej, Philip Scheltensj, Rik Vandenberghek, Silvy Gabelk,l, Karen Meersmansk,l, Sebastiaan Engelborghsm,n,o,p, Ellen E. De Roeckm,n,o, Kristel Sleegerso,q, Giovanni B. Frisonir,s, Olivier Blint, Jill C. Richardsonu, Regis Bordetv, Jose L. Molinuevow,x, Lorena Ramix, Anders Walliny, Petronella Kettuneny,z, Magda Tsolakiaa, Frans Verheyi, Alberto Lle obb, Daniel Alcoleacc, Julius Poppdd,ee, Gwendoline Peyratoutdd, Pablo Martinez-Lageff, Mikel Taintaff, Peter Johannsengg, Charlotte E. Teunissenhh, Yvonne Freund-Leviii,jj,kk, Lutz Fr€olichll, Cristina Legido-Quigleymm,nn, Frederik Barkhofoo,pp, Kaj Blennowe,qq, Henrik Zetterberge,qq,rr,ss, Susan Bakertt, B. Paul Morganh, Johannes Streffern,uu, Pieter Jelle Visseri,j,vv, Lars Bertramb,ww, Simon Lovestonea,xx, Alejo J. Nevado-Holgadoa a Department of Psychiatry, University of Oxford, Oxford, UK b L€ ubeck Interdisciplinary Platform for Genome Analytics, Institutes of Neurogenetics and Cardiogenetics, University of L€ubeck, L€ubeck, Germany c Institute of Clinical Molecular Biology, Christian-Albrechts-University of Kiel, Kiel, Germany d Maurice Wohl Clinical Neuroscience, Institute of Psychiatry, Psychology and Neuroscience, Kings College London, London, UK e Department of Psychiatry and Neurochemistry, University of Gothenburg, M€olndal, Sweden f Wallenberg Centre for Molecular & Translational Medicine, University of Gothenburg, Gothenburg, Sweden g NIHR Biomedical Research Centre for Mental Health and Biomedical Research Unit for Dementia at South London and Maudsley NHS Foundation, London, UK h Dementia Research Institute Cardiff, Cardiff University, Cardiff, UK i Department of Psychiatry and Neuropsychology, School for Mental Health and Neuroscience, Alzheimer Centrum Limburg, Maastricht University, Maastricht, the Netherlands j Alzheimer Center, VU University Medical Center, Amsterdam, the Netherlands k University Hospital Leuven, Leuven, Belgium l Laboratory for Cognitive Neurology, Department of Neurosciences, KU Leuven, Belgium m Center for Neurosciences, Vrije Universiteit Brussel (VUB), Brussels, Belgium n Reference Center for Biological Markers of Dementia (BIODEM), University of Antwerp, Antwerp, Belgium o Institute Born-Bunge, University of Antwerp, Antwerp, Belgium p Department of Neurology, VUB University Hospital Brussels (UZ Brussel), Brussels, Belgium S.L. has been named as an inventor on biomarker intellectual property suka and is the inventor of a patent on synaptic markers in CSF (all unrelated protected by Proteome Sciences and Kings College London; in both cases to this study). L.F. has received research funding, consultancy fees, and not the subject of the present study. He has served on the advisory board speech honoraria from Allergan, Avid-Eli Lilly, Avanir, Avraham Pharma- of SomaLogic Inc. and has provided consultancy for Merck and Eisai within ceuticals, Axon Neuroscience, Axovant, Biogen, Boehringer Ingelheim, Ei- the past 3 years. He is currently an employee of Janssen-Cilag, UK. H.Z. has sai, Functional Neuromodulation, GE Health Care, Lundbeck, Merck served at scientific advisory boards of Eli Lilly, Roche Diagnostics, Wave, Sharpe & Dohme, Novartis, Pfizer, Piramal Imaging, Roche, and Schwabe CogRx, and Samumed, has received travel support from Teva, and is a Pharma. J.P. has served at scientific advisory boards of Fujirebio Europe, Eli cofounder of Brain Biomarker Solutions in Gothenburg AB (all unrelated Lilly, and Nestle Institute of Health Sciences, all unrelated to this study. to this study). J.L.M. has served at scientific advisory boards of Eli Lilly, *Corresponding author. Tel.: 144 01865 618232; Fax: 144 01865 Roche Diagnostics, Roche, Biogen, Lundbeck, Novartis, IBL, Axovant, Or- 793101. yzon, and Merck (all unrelated to this study). A.L.B. has served at scientific E-mail address: liu.shi@psych.ox.ac.uk advisory boards of Fujirebio Europe, Eli Lilly, Novartis, Nutricia, and Ot- https://doi.org/10.1016/j.jalz.2019.06.4951 1552-5260/ Ó 2019 The Authors. Published by Elsevier Inc. on behalf of the Alzheimer’s Association. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). 15525279, 2019, 11, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1016/j.jalz.2019.06.4951 by Readcube (Labtiva Inc.), Wiley Online Library on [20/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License L. Shi et al. / Alzheimer’s & Dementia 15 (2019) 1478-1488 1479 q Neurodegenerative Brain Diseases Group, Center for Molecular Neurology, VIB, Edegem, Belgium r University of Geneva, Geneva, Switzerland s IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli, Brescia, Italy t AIX Marseille University, INS, Ap-hm, Marseille, France u Neurosciences Therapeutic Area, GlaxoSmithKline R&D, Stevenage, UK v University of Lille, Inserm, CHU Lille, Lille, France w Alzheimer’s disease & other cognitive disorders unit, Hopsital Clınic-IDIBAPS, Barcelona, Spain x Barcelona Beta Brain Research Center, Universitat Pompeu Fabra, Barcelona, Spain y Institute of Neuroscience and Physiology, Sahlgrenska Academy at University of Gothenburg, Gothenburg, Sweden z Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, UK aa 1st Department of Neurology, AHEPA University Hospital, Makedonia, Thessaloniki, Greece bb Hospital de la Santa Creu i Sant Pau, Barcelona, Spain cc Department of Neurology, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain dd University Hospital of Lausanne, Lausanne, Switzerland ee Geriatric Psychiatry, Department of Mental Health and Psychiatry, Geneva University Hospitals, Geneva, Switzerland ff CITA-Alzheimer Foundation, San Sebastian, Spain gg Danish Dementia Research Centre, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark hh Neurochemistry Laboratory, Department of Clinical Chemistry, Amsterdam Neuroscience, Amsterdam University Medical Centers, Vrije Universiteit, Amsterdam, The Netherlands ii 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 jj Department of Psychiatry in Region Orebro € € County and School of Medical Sciences, Faculty of Medicine and Health, Orebro € University, Orebro, Sweden kk Department of Old Age Psychiatry, Psychology and Neuroscience, Kings College London, London, UK ll Department of Geriatric Psychiatry, Zentralinstitut f€ur Seelische Gesundheit, University of Heidelberg, Mannheim, Germany mm Kings College London, London, UK nn The Systems Medicine Group, Steno Diabetes Center, Gentofte, Denmark oo Department of Radiology and Nuclear Medicine, VU University Medical Center, Amsterdam, The Netherland pp UCL Institutes of Neurology and Healthcare Engineering, London, UK qq Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, M€olndal, Sweden rr UK Dementia Research Institute at UCL, London, UK ss Department of Neurodegenerative Disease, UCL Institute of Neurology, London, UK tt Janssen R&D, Titusville, NJ, USA uu Janssen R&D, LLC, Beerse, Belgium vv Department of Neurobiology, Care Sciences and Society, Division of Neurogeriatrics, Karolinska Institutet, Stockholm, Sweden ww Department of Psychology, University of Oslo, Oslo, Norway xx Janssen-Cilag UK, formerly Department of Psychiatry, University of Oxford, Oxford, UK Abstract Introduction: Plasma proteins have been widely studied as candidate biomarkers to predict brain amyloid deposition to increase recruitment efficiency in secondary prevention clinical trials for Alz- heimer’s disease. Most such biomarker studies are targeted to specific proteins or are biased toward high abundant proteins. Methods: 4001 plasma proteins were measured in two groups of participants (discovery group 5 516, replication group 5 365) selected from the European Medical Information Framework for Alzheimer’s disease Multimodal Biomarker Discovery study, all of whom had measures of am- yloid. Results: A panel of proteins (n 5 44), along with age and apolipoprotein E (APOE) ε4, predicted brain amyloid deposition with good performance in both the discovery group (area under the curve 5 0.78) and the replication group (area under the curve 5 0.68). Furthermore, a causal relation- ship between amyloid and tau was confirmed by Mendelian randomization. Discussion: The results suggest that high-dimensional plasma protein testing could be a useful and reproducible approach for measuring brain amyloid deposition. Ó 2019 The Authors. Published by Elsevier Inc. on behalf of the Alzheimer’s Association. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Keywords: Amyloid b; SOMAscan assay; Plasma proteomics; Replication; Causal relationship; Tau 1. Introduction such trials present clinically with AD but lack amyloidosis [2]. Furthermore, most such trials have been conducted rela- Antiamyloid clinical trials for Alzheimer’s disease (AD) tively late in the disease process and targeting treatment to have been unsuccessful to date [1]. The failure of such trials earlier presymptomatic or prodromal stages of the disease was partially caused by the fact that participants enrolled in might have more success [3]. As a consequence, it has 15525279, 2019, 11, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1016/j.jalz.2019.06.4951 by Readcube (Labtiva Inc.), Wiley Online Library on [20/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1480 L. Shi et al. / Alzheimer’s & Dementia 15 (2019) 1478-1488 increasingly common to conduct clinical trials in preclinical program seeking to make data, and in these case samples, re- or prodromal disease where AD pathology is confirmed us- useable for secondary studies of neurodegeneration. The ing biomarkers [2,4]. Currently, the best characterized design of the EMIF-AD MBD study has been described pre- methods for measuring amyloid pathology are positron viously [29]. Briefly, 1221 participants meeting specified emission tomography (PET) imaging measures of brain and previously reported inclusion criteria were identified amyloid deposition and cerebrospinal fluid (CSF) from 11 preexisting European cohorts (“parent cohorts”) us- measurement of amyloid b (Ab) levels [5]. However, these ing the EMIF catalog (https://emif-catalogue.eu) and associ- measures are challenging because of invasiveness, cost, ated tools. Each parent cohort was approved by local and restricted availability [6,7]. medical ethics committees. As reported previously [29], a Blood-based biomarkers have therefore been investigated critical criterion for inclusion was that a measure of brain as a less-invasive and potentially cost-effective option for amyloid load was available, using either CSF Ab or amyloid early detection and monitoring of AD pathology. An PET scan. The classification of amyloid burden (high/low) in increasing number of studies [8–12], including those by the EMIF-AD MBD cohort has been described previously ourselves [13–19], have found that a range of proteins in [29]. Furthermore, Ab measurements were combined into plasma can reflect neocortical Ab burden and hence might a continuous variable using z-scoring, using the mean and act as biomarkers. Recent studies have added to these by standard deviation of the control subjects as a reference [29]. demonstrating that measures of Ab in blood can predict In addition to the primary outcome variable of amyloid brain amyloid with high accuracy [20–22], further load, another five markers in CSF were measured including indicating that blood proteins can be effective biomarkers total tau (t-tau), phosphorylated tau (p-tau), neurofilament for predicting brain pathology. light chain (NFL), neurogranin, and YKL-40 [29]. The The failure of antiamyloid clinical trials also provoked levels of t-tau and p-tau in CSF were measured locally, some to question the amyloid cascade hypothesis [23–25]. and the local cutoff point was used to determine their status Broadly speaking, the amyloid cascade hypothesis states (high/low). Consequently, the p-tau and t-tau values were Z- that over time, an imbalance in Ab production and/or scored with controls within each data set as a reference. The clearance leads to gradual accumulation and aggregation other three CSF markers (NFL, neurogranin, and YKL-40) of the peptide in the brain, initiating a neurodegenerative were measured in a central laboratory (Gothenburg Univer- cascade that involves not only amyloid deposition but also sity, Sweden), and their status (high/low) was determined by inflammation, tau pathology, neuronal dysfunction, and the median value of each marker within the whole data set. loss [26,27]. Despite some reservations regarding this In addition, the following AD-related phenotypes were hypothesis prompted by the failure of clinical trials, also measured: magnetic resonance imaging values of hip- amyloid still remains a central part of our understanding of pocampal volume; clinical assessments including baseline the pathophysiology of the disease. Nonetheless, assessing diagnosis, baseline Mini–Mental State Examination the relationship between amyloid and other AD (MMSE) score, and mild cognitive impairment (MCI) con- pathologies has become increasingly important. version; and finally, apolipoprotein E (APOE) and genome- With this in mind, we embarked on the study reported wide single nucleotide polymorphism (SNP) genotyping here, with two main objectives; first to identify a plasma mo- [29]. SNP assays were conducted using the “Global lecular signature of amyloid pathology for potential use as a Screening Array” (Illumina, Inc.) using standard procedures biomarker for preclinical or prodromal detection of disease from 250 ng of DNA extracted from whole blood [30]. Raw and second to investigate the causal relationship between data processing and initial quality control (QC) were per- amyloid and other AD pathologies. We report here a repli- formed in GenomeStudio (v2.0.2) using GenTrain (v3.0) cating signature of disease reflecting brain amyloid load as clustering algorithm generating genotypes for 645,896 and other features of AD, including tau pathology, in blood. SNPs. Postprocessing was performed in PLINK (v1.9) using Then using Mendelian randomization (MR) [28], an standard QC thresholds (i.e., sample genotyping efficiency approach more often used in genetic studies, we find a causal . 95%, SNP genotyping efficiency . 98%, departures relationship between amyloid and tau while such reverse from Hardy Weinberg equilibrium at P value , .000005, relationship was not found. and a minor allele frequency [MAF] of , 0.01). Eventually, this left 490,717 SNPs for analysis. Genotypes for recon- struction of the APOE ε2/3/4 haplotypes were either gener- 2. Methods ated on the Global Screening Array (rs7412) or by manual genotyping (rs429358) as described previously [31]. We 2.1. Participants: EMIF-AD Multimodal Biomarker classified individuals as APOE ε4 carriers or noncarriers ac- Discovery study cording to their genotype status at rs429358 (C-allele 5 ε4). The EMIF-AD Multimodal Biomarker Discovery (MBD) The 1221 subjects selected for the EMIF-AD MBD study is part of the European Medical Information Frame- cohort included 492 healthy controls (HCs), 527 MCI, and work for Alzheimer’s disease (EMIF-AD; http://www. 202 AD participants. Among these participants, we included emif.eu); a public-private precompetitive cross European in the study all those that had plasma samples available (total 15525279, 2019, 11, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1016/j.jalz.2019.06.4951 by Readcube (Labtiva Inc.), Wiley Online Library on [20/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License L. Shi et al. / Alzheimer’s & Dementia 15 (2019) 1478-1488 1481 881). These samples were then divided into two separated features that produced best performance were selected from groups, with each group of samples being processed inde- group 1. To verify replication, the optimal classifier was pendently. Group 1 was used as a discovery group and then validated in the second and combined cohorts for distin- included 150 HCs (29.1%), 188 MCI (36.4%), and 178 guishing Ab status and other AD endophenotypes. AD (34.5%). Group 2 was used for validation and included 161 HCs (44.1%), 198 MCI (54.2%), and 6 AD (1.6%). 2.5. Mendelian randomization Furthermore, we sought to maintain amyloid balance in con- trols and MCI individuals given that the objective is to iden- MR was used to investigate the causal relationship be- tify biomarkers of amyloid pathology in preclinical stage. As tween amyloid and tau pathology. MR uses genetic variants a result, in this combined group of samples from people in as an instrument when they are robustly associated with a the HC and MCI categories (total n 5 697), there were risk factor (exposure), with the fundamental assumption 338 with and 359 without amyloid pathology. that they are associated with the studied outcome through that exposure. As genetic variants cannot be altered and 2.2. SOMAscan assay are static throughout an individual’s lifetime, reverse causa- tion can be excluded, rendering MR a powerful tool to We measured proteins in plasma using the SOMAscan examine causality between the exposure and outcome [28]. assay platform (SomaLogic Inc.). SOMAscan is an To explore the causal relationship between amyloid and tau aptamer-based assay allowing for the simultaneous measure- (both t-tau and p-tau), we first performed a systematic review ment and quantification of, in the version used here, 4001 of the literature to search SNPs that were significantly associ- proteins. The assay uses chemically modified nucleotides ated with amyloid but not with tau or p-tau. Overall, 78 SNPs to transform a protein signal into a nucleotide signal that were found to be significantly associated with abnormal amy- can be quantified using relative fluorescence on microarrays loid pathology measured by PET scan, CSF amyloid, or post- [32]. As noted previously, groups of samples were measured mortem measurement (Supplementary Table 1). Six SNPs independently. Forty subjects were tested in both batches to were excluded including 3 which were also associated with normalize the data across assay runs. t-tau and p-tau and 3 which were within 1 Mb of APOE (i.e., APOC1 and TOMM40), given that APOE is strongly 2.3. Statistical analysis associated with both amyloid and tau. Of the remaining All statistical analyses were completed using R (version SNPs, 6 were not measured using the assay we used, leaving 3.3.2). The Mann-Whitney U test was used to compare par- 66 SNPs for further analysis (Supplementary Table 1). The as- ticipants’ characteristics between high- and low-amyloid cat- sociation of these 66 SNPs with the Z-score of amyloid, t-tau, egories in both groups. The association between proteins and and p-tau was calculated using PLINK (v1.7). Finally, based amyloid Z-score was calculated by partial Spearman correla- on the association, we used four MR approaches, including tion, adjusting for five covariates including age, sex, APOE MR Egger linear regression, weighted median, inverse- ε4 status, study of origin, and number of blood freeze-thaw variance weighted (IVW), and simple median [33], to investi- cycles. False discovery rate correction was used to correct gate the causal relationship between amyloid and tau (both P values for multiple comparisons. Proteins that were differ- t-tau and p-tau). Each MR method is based on different as- entially expressed at a significance level of P , .05 were sumptions. The Egger method is sensitive to SNP pleiotropy further nominated for pathway analysis using the Metascape and allows the estimation of underlying bias by allowing a software (http://metascape.org/gp/index.html#/main/step1). non-zero estimate for the intercept of the calculated ratio of Briefly, differentially expressed proteins were inputted as b values [34]. The weighted median method uses a median “protein list” and all 4001 proteins measured by SOMAscan of the individual causal estimate per SNP, which is calculated assay were used as “background.” This enrichment analysis from the ratio estimates of outcome’s regression coefficient was performed on the KEGG database. divided by exposure [35]. The IVW method uses the same ra- tio estimates but incorporates IVWs into the final summary es- timate [36]. Comparing estimates from all the methods shows 2.4. Machine learning the robustness of the overall analysis. Furthermore, to deter- Machine learning was used to identify optimal multivariate mine if there was any single SNP driving the relationship, signatures, including both proteins and demographic data (age we performed a leave-one-out analysis where the MR is per- and APOE ε4) as input features, to differentiate between high formed removing a different SNP in each iteration [37]. and low Ab status in group 1. First, the effects of study and A similar approach was performed to investigate the blood freeze-thaw cycle on proteins were removed by linear causal relationship between t-tau and amyloid. Using the regression and the residuals were used for analysis. Subse- same approach, we identified 21 SNPs that were signifi- quently, Lasso was used to select the “n” top input features cantly associated with the abnormal CSF t-tau level but that best differentiated high from low Ab. Support vector ma- not with amyloid (Supplementary Table 2). After checking chine classifiers were then built using from 1 to 100 features to the effect allele, eighteen SNPs passed QC and were used predict the outcome under 10-fold cross-validation. Those for MR analysis (Supplementary Table 2). The causal 15525279, 2019, 11, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1016/j.jalz.2019.06.4951 by Readcube (Labtiva Inc.), Wiley Online Library on [20/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1482 L. Shi et al. / Alzheimer’s & Dementia 15 (2019) 1478-1488 Table 1 Demographics of the study population Group 1 (n 5 516) Group 2 (n 5 365) Characteristics High Ab (n 5 369) Low Ab (n 5 147) P value High Ab (n 5 134) Low Ab (n 5 231) P value Ab Z-score 21.39 6 0.50 0.49 6 0.57 ,.001* 21.22 6 0.40 0.52 6 0.67 ,.001* Age (yrs) 70.0 6 8.2 68.3 6 8.4 .04* 70.0 6 8.3 64.9 6 8.1 ,.001* Female sex, N (%) 210 (57) 62 (42) .002* 70 (52) 139 (60) .14 Male sex, N (%) 159 (43) 85 (58) .002* 64 (48) 92 (40) .14 APOE ε41, N (%) 244 (66) 43 (29) ,.001* 76 (57) 53 (23) ,.001* APOE ε42, N (%) 125 (34) 104 (71) ,.001* 58 (43) 178 (77) ,.001* MMSE 24.1 6 4.6 27.0 6 3.2 ,.001* 26.9 6 2.7 28.2 6 1.8 ,.001* NOTE. Percentage of cases is shown in brackets for female and male sex, as well as APOE ε4 carriers and noncarriers in each amyloid b (Ab) category. P values compare each demographic across the high Ab and low Ab groups. Group 1 containing 516 participants was used for discovery, and group 2 of 365 participants for validation. Abbreviations: APOE, apolipoprotein E; MMSE, Mini–Mental State Examination. *P , .05. relationship between p-tau and amyloid was not presented identify prodromal AD, and in AD itself, to add to the accu- because few SNPs were found to be strongly associated racy of diagnosis. with p-tau. Therefore, this approach was underpowered for Overall, 3676 and 3916 proteins passed QC measures in making causal inferences. group 1 and group 2, respectively. Using partial correlation to analyze the association between proteins and amyloid Z-score, we found that in group 1, 301 proteins reached sta- 3. Results tistical significance (P , .05) and 5 of them reached false 3.1. Subject demographics discovery rate (q , 0.1) (Supplementary Table 3), corrected for multiple comparisons. In group 2, 536 proteins reached The EMIF-MBD cohort comprises 1221 subjects of statistical significance (P , .05) while none passed false dis- which 881 had samples available for this study. As described covery rate. Fig. 1A and B shows volcano plots of proteins in the Methods section, these were divided into two groups from both groups with cutoff P 5 .05. for discovery and validation purposes. Table 1 shows the We then merged the lists of differentially expressed pro- characteristics of the participants from two groups split by teins obtained from both groups finding a total of 786 pro- amyloid status (high/low). For both groups, the participants teins associated with high versus low amyloid. This with high amyloid were older (P 5 .04 in group 1, P , .001 number of dysregulated proteins was significantly higher in group 2), the MMSE was lower (P , .001 in group 1, than that obtained by chance alone (P , .001). We then P , .001 in group 2), and the prevalence of APOE ε4 used the Metascape software, performed on the KEGG data- carriers was higher (P , .001 in group 1, P , .001 in group base, to assess the biological significance of these dysregu- 2). A significant difference in sex was found in group 1 lated proteins. Pathway analysis of the 786 differentially (P 5 .002), whereas there was no difference in group 2 expressed proteins revealed 10 significantly enriched path- (P 5 .14). ways: arginine biosynthesis (P 5 .003), inflammatory medi- ator regulation of TRP channels (P 5 .007), 3.2. Proteins significantly associated with amyloid glycerophospholipid metabolism, (P 5 .010), citrate cycle Z-score in both groups (P 5 .012), Notch signaling (P 5 .016), RNA degradation (P 5 .023), type II diabetes mellitus (P 5 .038), peroxisome In this study, we sought to find plasma biomarkers related (P 5 .038), fatty acid degradation (P 5 .041), and NOD-like to amyloid pathology independent of diagnosis and so com- receptor signaling (P 5 .043) pathways. bined samples from study participants across the AD diag- nostic spectrum (HC, MCI, and AD) and used measures of 3.3. Multiprotein classifier of amyloid normal/abnormal amyloid pathology as the principle outcome measure. status and other AD endophenotypes Combining samples in this way reflects the operational need for a biomarker predicting brain amyloid pathology. Having demonstrated that plasma proteins are signifi- Recent experience suggests that up to 20% of people with cantly altered in samples from people with abnormal Ab a diagnosis of AD in clinical trials had little or no brain am- levels, we then sought to find a minimal signal from these yloid pathology, demonstrating that clinical classification is plasma data that might act as a biomarker indicative of brain an inadequate predictor of brain pathology. Moreover, a amyloid pathology. To do this we performed machine biomarker of brain pathology in all three states—HC, learning with 10-fold cross-validation in group 1 to identify MCI, and AD—would have value. In HC, a marker of brain the optimal multivariate signatures that differentiated be- amyloid would help to identify preclinical AD, in MCI, to tween high- and low-Ab individuals. We first used 15525279, 2019, 11, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1016/j.jalz.2019.06.4951 by Readcube (Labtiva Inc.), Wiley Online Library on [20/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License L. Shi et al. / Alzheimer’s & Dementia 15 (2019) 1478-1488 1483 Fig. 1. Correlation of plasma proteins with amyloid Z-score by partial Spearman correlation. (A and B) Volcano plot of proteins obtained from group 1 and group 2, respectively. demographic variables (age, sex, and education) with APOE inating AD from HC and AD from MCI, differentiating sta- ε4 as input features to predict amyloid burden. Results tus of t-tau, p-tau, MCI conversion, baseline MMSE score, showed that a combination of age and APOE ε4 achieved hippocampal volume, and other three CSF biomarkers the highest predictive value AUC of 0.70. Then, we added including NFL, neurogranin, and YKL-40. As shown in all measured proteins to these demographic variables and Fig. 2B, the panel predicted most of these AD phenotypes we found that a panel of 46 features achieved the highest pre- with an AUC of .0.6, except for CSF NFL, neurogranin, dictive value AUC of 0.78. The input features automatically and YKL-40. The sample size and AUC of each AD pheno- selected by the classifier were age, APOE ε4, and 44 proteins type is shown in Supplementary Table 5. (Supplementary Table 4). The AUC in training and testing sets for different number of input features is shown in 3.4. Causal relationship between amyloid and tau Supplementary Fig. 1. We then investigated the performance pathology of this classifier on differentiating amyloid status in the sec- ond and independent group, maintaining the parameters ob- We then sought to use these data to address the critically tained from group 1. The results showed that the same panel important question of causality in relation to AD patholog- achieved an AUC of 0.68, demonstrating good replication ical features. To investigate the causal relationship between (Fig. 2A). amyloid and tau, we first identified SNPs associated with We further investigated the performance of the same amyloid pathology from systematic review of the literature. panel to predict other AD endophenotypes in the combined We found 72 SNPs (Supplementary Table 1) significantly cohort (combining groups one and two), including discrim- associated with abnormal amyloid but not with tau Fig. 2. (A) AUC of the 46 signatures to differentiate amyloid status in group 1 and group 2. (B) AUC of the 46 signatures to differentiate other AD endophe- notypes in the combined cohort. Abbreviations: AUC, area under the curve; HC, healthy control; MCI, mild cognitive impairment; t-tau, total tau; p-tau, phos- phorylated tau; MMSE, Mini–Mental State Examination; NFL, neurofilament light chain. 15525279, 2019, 11, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1016/j.jalz.2019.06.4951 by Readcube (Labtiva Inc.), Wiley Online Library on [20/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1484 L. Shi et al. / Alzheimer’s & Dementia 15 (2019) 1478-1488 Table 2 relationship was found between t-tau and neurogranin as MR estimates of the causal effect of amyloid on t-tau and p-tau as well as the well as between t-tau and YKL-40 (Supplementary Table 6). causal effect of t-tau on amyloid Study type Methods b se 95% CI P value 4. Discussion Amyloid MR Egger 0.85 0.15 0.54 to 1.15 ,.001 SNPs with t-tau Weighted 0.73 0.22 0.31 to 1.16 ,.001 In this study, we used the SOMAscan assay to measure median 4001 proteins in plasma in two groups of subjects from the IVW 0.86 0.11 0.65 to 1.08 ,.001 Simple 0.81 0.50 20.17 to 1.80 .110 EMIF-AD MBD study. Using machine learning, we identi- median fied a panel of 44 proteins that, along with age and APOE Amyloid MR Egger 0.73 0.17 0.39 to 1.06 ,.001 ε4, predicted measures of central amyloid with good perfor- SNPs with p-tau Weighted 0.62 0.23 0.17 to 1.06 .006 mance in both the discovery group (AUC 5 0.78) and the median replication group (AUC 5 0.68). The same panel also pre- IVW 0.86 0.12 0.62 to 1.10 ,.001 Simple 1.31 0.52 0.29 to 2.33 .014 dicted other AD phenotypes, including CSF t-tau, CSF median p-tau, MCI conversion, baseline MMSE score, and hippo- t-tau MR Egger 20.45 0.36 21.15 to 0.26 .23 campal volume. Moreover, a causal relationship between SNPs with Weighted 20.31 0.29 20.87 to 0.25 .27 amyloid and t-tau and with p-tau was confirmed using a amyloid median MR approach, whereas evidence for the reverse relationship IVW 20.02 0.20 20.42 to 0.38 .93 Simple 20.20 2.05 24.22 to 3.81 .92 between t-tau and amyloid was not found. median These results are in concordance with one of our previous studies [18], in which we used the same SOMAscan assay to Abbreviations: b, b coefficient; CI, confidence interval; IVW, inverse- variance weighted; MR, Mendelian randomization; t-tau, total tau; p-tau, measure plasma proteins in 58 cognitively healthy men. In phosphorylated tau; SNP, single nucleotide polymorphism. this previous study, 667 proteins were significantly associ- ated with CSF amyloid measurement, and of these 167 pro- pathology of which 66 were assayed using the array used in teins (25%) overlapped with the differentially expressed this study (Supplementary Table 1) and hence available for proteins obtained in the present study. Furthermore, the MR analysis. Results showed that amyloid was associated number of overlapping proteins was significantly higher with a moderate alteration in t-tau when using three MR ap- than expected from chance alone (P 5 .01), confirming proaches including the Egger method (b 5 0.85, se 5 0.15, that the SOMAscan assay may be a useful and powerful 95% CI [0.54, 1.15], P , .001), the IVW method (b 5 0.86, screening tool for identifying plasma proteins reflecting am- standard error of the effect size [se] 5 0.11, 95% CI [0.65, yloid deposition in early, even preclinical, phases of disease 1.08], P , .001), and weighted median method (b 5 0.73, and across the clinical AD spectrum as in the present study. se 5 0.22, 95% CI [0.31, 1.16], P , .001), though not Pathway analysis of the 786 differentially expressed pro- from simple median method (b 5 0.81, se 5 0.50, 95% CI teins revealed 10 significantly enriched pathways. When [-0.17, 1.80], P 5 .11; Table 2, Supplementary Fig. 2). comparing these enriched pathways with AlzPathway, a Altered amyloid was also associated with p-tau in all four comprehensive map of AD signaling and related pathways MR approaches (Table 2, Supplementary Fig. 2). Results [38,39], we found two pathways common to both; Notch from leave-one-out analysis demonstrated that no single signaling and type II diabetes mellitus. This finding is SNP was driving the majority of the association signal be- interesting given that multiple lines of evidence have tween amyloid and either t-tau or p-tau (Supplementary previously shown that Notch signaling and type II diabetes Fig. 3). In addition, we explored the causal relationship be- mellitus are associated with AD [40–43]. Our results add tween amyloid and other three CSF markers including NFL, weight to this finding and indicate that proteins involved in neurogranin, and YKL-40; no significant association be- these two pathways could be treatment targets of AD. tween amyloid and these three markers was found From the discovery phase of the present study, a panel of (Supplementary Table 6). 46 features achieved a high AUC in discriminating high from We then reversed the direction of analysis and investi- low brain amyloid (Fig. 2A). Age and APOE ε4 contributed gated whether there was a causal relationship between t- to the model, as in previous studies seeking biomarkers tau and amyloid. Twenty-one SNPs (Supplementary indicative of brain amyloid; unsurprisingly given that both Table 2) were found to be significantly associated with the age and APOE genotype are highly associated with amyloid abnormal CSF t-tau level from systematic review, among pathology [44–48]. Despite this, the protein signature which eighteen SNPs (Supplementary Table 2) passed QC consisting of 44 proteins achieved a greater discrimination and were used for MR analysis. Results showed no causal performance in the discovery group than age and APOE ε4 relationship between t-tau and amyloid from any of the (0.78 vs. 0.70). In this study, as the sample collection was four MR approaches (Table 2). We also investigated the deliberately selected to overrepresent high amyloid causal relationship between t-tau and other three CSF individuals, the proportion of APOE ε4 carriers in the markers (NFL, neurogranin, and YKL-40). No causal rela- sample collection is higher than in the general population tionship between t-tau and NFL was found, whereas such (47.2% vs. 10-20% [49]). In the real world of clinical trials 15525279, 2019, 11, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1016/j.jalz.2019.06.4951 by Readcube (Labtiva Inc.), Wiley Online Library on [20/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License L. Shi et al. / Alzheimer’s & Dementia 15 (2019) 1478-1488 1485 and indeed clinical practice, APOE is of less value in identi- prescreening processes to recruit people to clinical trials fying people with likely AD pathology as most people in the where screening includes determination of amyloid status early stages of disease are APOE ε4 negative. Because of by CSF or PET markers. Currently, the very high screen fail- this, an algorithm that relies on APOE genotype to identify ure rate at this stage slows clinical development and adds people for inclusion in clinical trials is of limited value, substantially to the cost of clinical trials while also exposing and hence, the identification of additional markers, such as large numbers of potential participants to relatively invasive we report here, is even more valuable than a modest increase tests. Our data suggest that this prescreen failure when in AUC over a predictor that includes APOE would suggest. seeking people with brain pathology could be reduced using This study is the largest we are aware of to report a plasma a simple blood assay. In future years, an assay such as this biomarker indicative of AD pathology both in terms of the could be used as part of a clinical workup directing at risk number of proteins assayed and in samples size. Recently, individuals to specific testing—such as by CSF measures Nakamura et al. (2018) [20] and Ovod et al. (2017) [21] re- or using PET imaging—before therapeutic intervention. ported methods to measure plasma Ab directly, and Ashton In addition to predicting amyloid status, the 44 proteins et al. (2019) [50] used high-resolution mass spectrometry to along with age and APOE ε4 could also predict other AD en- identify plasma proteins that correlate with altered central dophenotypes including discriminating AD from HC and amyloid. In all three cases, these assays achieve higher mea- AD from MCI, differentiating status of t-tau, p-tau, MCI sures of accuracy than we report. However, our study is conversion, baseline MMSE score, and hippocampal vol- considerably larger and conducted in a mixed cohort derived ume. These findings are in concordance with some recent from multiple different sources. By virtue of this size and the studies [52,53] and further demonstrate the association of likely heterogeneity of the sample characteristics, it is amyloid pathology with other pathologies of AD. possible that the results we report here are likely to withstand Moreover, these findings are potentially useful in clinical the rigors of application in the real-world situations of multi- trial settings especially as increasing number of clinical centre, multinational complex clinical trials. Moreover, the trials are directed at tau-related targets [1]. The biomarker analytical platform we use here is clinic-ready and relatively panel that we report here could potentially be used in very low cost. Finally, using the panel we describe could have sig- large numbers of people to identify those likely to have nificant economic benefits. Based on the performance of our both amyloid and tau pathology, even in preclinical phase. panel (AUC 5 0.78, sensitivity 5 0.84, specificity 5 0.54), Despite more than two decades of study, the mechanism of we obtain a likelihood ratio of 1.82, suggesting that the panel AD pathology is still disputed with serial trials failure provok- could improve approximately 15% in amyloid detection [51] ing some to question the amyloid cascade hypothesis. Broadly relative to current practice. Given that screen failure rates for speaking, this states that the generation of Ab leads, through a amyloid positive recruitment to clinical trials is approxi- series of largely unknown steps, to tau pathology and hence mately 70%, then to recruit 100 individuals there is a need neuronal dysfunction and dementia [26,27]. Using the data to enroll 333 participants with a cost of approximately generated here we confirmed the relationship between $1.7 million ($5000 per scan). A 15% improvement as we amyloid and tau pathology using two approaches. First, the demonstrate would reduce the number needed to screen proteins selected from amyloid prediction also predict tau from 333 to 222, with a cost saving of $0.55 million per pathology (CSF t-tau and p-tau). Second, we used MR, a 100 participants of a clinical trial. Therefore, as trials typi- method more commonly used in genetic studies to go cally seek to recruit large numbers of participants, especially beyond association to causality, finding that higher amyloid in late phase, we believe that despite the moderate improve- levels drive a moderate increase in both t-tau and p-tau but ment of AUC, these 44 proteins obtained from our large sam- that the reverse (t-tau driving amyloid pathology) is not true. ple size study are strong candidates and worthy of further Our results confirm the crucial role of amyloid in driving tau investigation. pathology marking the onset of preclinical AD, and The National Biomarker Development Alliance proposes consistent with previous findings [53–55]. 6 steps for any biomarker development (nbdabiomarkers. Interestingly, no such association was found between amy- org/) including early discovery, translatable discovery, assay loid and other three CSF biomarkers (NFL, neurogranin, and development, assay performance, biomarker quantification, YKL-40), either from using proteins to discriminate their and biomarker validation. Here, we have discovered and levels or from MR analysis. This suggests that these three bio- replicated 44 nominated biomarkers, the subsequent phase markers (NFL, neurogranin, and YKL-40) play different roles of research will be “translatable discovery,” indicating that in AD independent of, or at least remote from, amyloid pathol- we need to replicate these 44 proteins in even larger indepen- ogy. Although their role is not understood completely, changes dent cohorts to further minimize overfitting and to obtain a in NFL are thought to reflect axonal damage [56], changes in smaller panel of proteins. One of the advantages of the plat- neurogranin to reflect synaptic dysfunction [57], and changes form that we used here, SOMAscan, is that it is a high- in YKL-40 to provide a measure of neuroinflammation quality assay platform already in increasingly wide research [58,59]. Our data suggest that amyloid is strongly and use and appropriate for clinical implementation. In the short causally associated with tau pathology but not with synaptic to medium term, the results we present here might be used in dysfunction or neuroinflammation. Instead, we found that 15525279, 2019, 11, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1016/j.jalz.2019.06.4951 by Readcube (Labtiva Inc.), Wiley Online Library on [20/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1486 L. Shi et al. / Alzheimer’s & Dementia 15 (2019) 1478-1488 tau is causally associated with neurogranin and YKL-40, indi- Center for Molecular Neurology. The research at VIB- cating that synaptic dysfunction and neuroinflammation are CMN is funded in part by the University of Antwerp consequences of tau-pathological processes. Interestingly, Research Fund. F.B. is supported by the NIHR biomedical similar observations have been made and are in line with our research centre at UCLH. L.S. is funded by DPUK through findings. For example, in multiple transgenic models, amyloid MRC (grant no. MR/L023784/2) and the UK Medical pathology results in very little neuronal damage, synaptic loss, Research Council Award to the University of Oxford (grant or indeed neuroinflammation, and where such pathological no. MC_PC_17215). consequences do exist, they are abolished in the context of Authors’ contributions: S.W., A.L.B., and S.L. contributed ablation of tau [60–62]. Although these findings are to study concept and design, and all authors contributed to necessarily preliminary, they suggest that proteomic and sample collection/selection and/or interpretation of data. perhaps other high-dimensionality molecular biomarker L.S. and A.J.N.-H. carried out data analysis and interpreta- studies might also contribute to an understanding of patholog- tion. L.S. drafted the manuscript, and all authors revised ical processes in disease. the manuscript. We acknowledge that the sample origin is a limitation of Ethics statement: Written informed consent was obtained this study. Given that both the discovery and replication from all participants before inclusion in the study. The med- samples are taken from the same meta-sample collection, ical ethics committee at each site approved the study. it is not an independent validation in the traditional sense. Data availability: The data sets generated and analyzed However, because the EMIF-AD MBD study is a multi- during the present study are available via the EMIF-AD centre study and collected samples from 11 centers across Catalogue via submitted research questions, which have Europe, the meta-collection reflects the challenges faced in to be approved by the data-owners from each parent the real world of multisite and multinational clinical cohort studies. Therefore, the replication obtained in this study in such a collection of samples has a higher probability Supplementary Data of effective utility in practice. In conclusion, our results strongly suggest that testing Supplementary data related to this article can be found at https://doi.org/10.1016/j.jalz.2019.06.4951. high-dimensional plasma protein levels is a useful and repro- ducible approach to measure central nervous system amy- loid deposition. This can be potentially applied as a prescreen to preselect patients for further selection proced- RESEARCH IN CONTEXT ures for clinical trials, thus reducing the cost incurred to clin- ical trials by screen failure. Furthermore, we confirmed the causal relationship between amyloid and tau pathology 1. Systematic review: We reviewed the literature using from MR. PubMed sources, meeting abstracts, and presenta- tions. Plasma proteins have been widely studied as Acknowledgments candidate biomarkers to predict brain amyloid depo- sition, but most such studies are targeted to specific This research was conducted as part of the EMIF-AD proj- proteins or are biased toward high abundant proteins. ect, which has received support from the Innovative Medi- Furthermore, the mechanism of AD pathology is still cines Initiative Joint Undertaking under EMIF grant disputed, with serial trials failure provoking some to agreement no. 115372, resources of which are composed question the amyloid cascade hypothesis. of financial contribution from the European Union’s Seventh Framework Program (FP7/2007-2013), and EFPIA com- 2. Interpretation: Our findings showed that testing panies’ in-kind contribution. The DESCRIPA study was plasma proteins using the high-dimensionality SO- funded by the European Commission within the 5th frame- MAscan assay could be a useful and reproducible work program (QLRT-2001-2455). The EDAR study was approach for indicating brain amyloid deposition as funded by the European Commission within the 5th frame- well as tau pathology. Furthermore, we confirmed the work program (contract # 37670). The Leuven cohort was crucial role of amyloid pathology in AD process with funded by the Stichting voor Alzheimer Onderzoek (grant a novel use of Mendelian randomization. numbers #11020, #13007, and #15005). R.V. is a senior clin- 3. Future directions: Although it looks promising that a ical investigator of the Flemish Research Foundation blood test could be used as a tool to preselect individ- (FWO). J.S. is currently an employee of UCB, Braine-l’Al- uals for further clinical trials, the plasma biomarkers leud, Belgium. The San Sebastian GAP study is partially identified from this study need to be further validated funded by the Department of Health of the Basque Govern- in a larger and independent cohort. ment (allocation 17.0.1.08.12.0000.2.454.01.41142.001.H). The authors acknowledge the contribution of the personnel of the Genomic Service Facility at the VIB-U Antwerp 15525279, 2019, 11, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1016/j.jalz.2019.06.4951 by Readcube (Labtiva Inc.), Wiley Online Library on [20/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License L. Shi et al. / Alzheimer’s & Dementia 15 (2019) 1478-1488 1487 References [22] Janelidze S, Stomrud E, Palmqvist S, Zetterberg H, van Westen D, Jeromin A, et al. Plasma beta-amyloid in Alzheimer’s disease and vascular disease. Sci Rep 2016;6:26801. [1] Cummings J, Lee G, Ritter A, Zhong K. Alzheimer’s disease drug [23] Herrup K. The case for rejecting the amyloid cascade hypothesis. Nat development pipeline: 2018. Alzheimers Dement (N Y) 2018; Neurosci 2015;18:794–799. 4:195–214. [24] Karran E, De Strooper B. The amyloid cascade hypothesis: are we [2] Cummings J. Lessons Learned from Alzheimer Disease: Clinical Tri- poised for success or failure? J Neurochem 2016;139:237–252. als with Negative Outcomes. Clin Transl Sci 2018;11:147–152. [25] Ricciarelli R, Fedele E. The Amyloid Cascade Hypothesis in Alz- [3] Golde TE, DeKosky ST, Galasko D. Alzheimer’s disease: The right heimer’s Disease: It’s Time to Change Our Mind. Curr Neuropharma- drug, the right time. Science 2018;362:1250–1251. col 2017;15:926–935. [4] Cummings JL. Biomarkers in Alzheimer’s disease drug development. [26] Barage SH, Sonawane KD. Amyloid cascade hypothesis: Pathogenesis Alzheimers Dement 2011;7:e13–44. and therapeutic strategies in Alzheimer’s disease. Neuropeptides [5] Cohen AD, Landau SM, Snitz BE, Klunk WE, Blennow K, 2015;52:1–18. Zetterberg H. Fluid and PET biomarkers for amyloid pathology in Alz- [27] Lemere CA, Masliah E. Can Alzheimer disease be prevented by heimer’s disease. Mol Cell Neurosci 2019;97:3–17. amyloid-beta immunotherapy? Nat Rev Neurol 2010;6:108–119. [6] de Almeida SM, Shumaker SD, LeBlanc SK, Delaney P, Marquie- [28] Grover S, Del Greco MF, Stein CM, Ziegler A. Mendelian Randomi- Beck J, Ueland S, et al. Incidence of post-dural puncture headache zation. Methods Mol Biol 2017;1666:581–628. in research volunteers. Headache 2011;51:1503–1510. [29] Bos I, Vos S, Vandenberghe R, Scheltens P, Engelborghs S, Frisoni G, [7] Lista S, Faltraco F, Prvulovic D, Hampel H. Blood and plasma-based et al. The EMIF-AD Multimodal Biomarker Discovery study: design, proteomic biomarker research in Alzheimer’s disease. Prog Neurobiol methods and cohort characteristics. Alzheimers Res Ther 2018;10:64. 2013;101–102:1–17. [30] Hong S, Dobricic V, Smith RG, K€uç€ukali F, Kilpert F, Bos I, et al. Ge- [8] Olsson B, Lautner R, Andreasson U, Ohrfelt A, Portelius E, Bjerke M, nomics and epigenomics analyses in the EMIF-AD multimodal et al. CSF and blood biomarkers for the diagnosis of Alzheimer’s dis- biomarker discovery study. Alzheimer’s Dement J Alzheimer’s Assoc ease: a systematic review and meta-analysis. Lancet Neurol 2016; 2018;14:P204. 15:673–684. [31] Ten Kate M, Redolfi A, Peira E, Bos I, Vos SJ, Vandenberghe R, et al. [9] Khan AT, Dobson RJ, Sattlecker M, Kiddle SJ. Alzheimer’s disease: MRI predictors of amyloid pathology: results from the EMIF-AD are blood and brain markers related? A systematic review. Ann Clin Multimodal Biomarker Discovery study. Alzheimers Res Ther 2018; Transl Neurol 2016;3:455–462. 10:100. [10] Carmona P, Molina M, Toledano A. Blood-based biomarkers of Alz- [32] Gold L, Ayers D, Bertino J, Bock C, Bock A, Brody EN, et al. Ap- heimer’s disease: diagnostic algorithms and new technologies. Curr tamer-based multiplexed proteomic technology for biomarker discov- Alzheimer Res 2016;13:450–464. ery. PLoS One 2010;5:e15004. [11] Baird AL, Westwood S, Lovestone S. Blood-Based Proteomic Bio- [33] Yavorska OO, Burgess S. Mendelian Randomization: an R package for markers of Alzheimer’s Disease Pathology. Front Neurol 2015;6:236. performing Mendelian randomization analyses using summarized [12] Thambisetty M, Lovestone S. Blood-based biomarkers of Alzheimer’s data. Int J Epidemiol 2017;46:1734–1739. disease: challenging but feasible. Biomark Med 2010;4:65–79. [34] Bowden J, Davey Smith G, Burgess S. Mendelian randomization with [13] Thambisetty M, Tripaldi R, Riddoch-Contreras J, Hye A, An Y, invalid instruments: effect estimation and bias detection through Egger Campbell J, et al. Proteome-based plasma markers of brain amyloid- regression. Int J Epidemiol 2015;44:512–525. beta deposition in non-demented older individuals. J Alzheimers Dis [35] Angrist JD, Imbens GW. Two-stage least squares estimation of average 2010;22:1099–1109. causal effects in models with variable treatment intensity. J Am Stat [14] Ashton NJ, Kiddle SJ, Graf J, Ward M, Baird AL, Hye A, et al. Blood Assoc 1995;90:431–442. protein predictors of brain amyloid for enrichment in clinical trials? [36] Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. A basic intro- Alzheimers Dement (Amst) 2015;1:48–60. duction to fixed-effect and random-effects models for meta-analysis. [15] Westwood S, Leoni E, Hye A, Lynham S, Khondoker MR, Ashton NJ, Res Synth Methods 2010;1:97–111. et al. Blood-based biomarker candidates of cerebral amyloid using PiB [37] Hemani G, Zheng J, Wade KH, Laurin C, Elsworth B, Burgess S, PET in non-demented elderly. J Alzheimers Dis 2016;52:561–572. et al. MR-Base: a platform for systematic causal inference across [16] Kiddle SJ, Thambisetty M, Simmons A, Riddoch-Contreras J, Hye A, the phenome using billions of genetic associations. BioRxiv Westman E, et al. Plasma based markers of [11C] PiB-PET brain am- 2016:78972. yloid burden. PLoS One 2012;7:e44260. [38] Ogishima S, Mizuno S, Kikuchi M, Miyashita A, Kuwano R, [17] Voyle N, Baker D, Burnham SC, Covin A, Zhang Z, Sangurdekar DP, Tanaka H, et al. AlzPathway, an updated map of curated signaling et al. Blood Protein Markers of Neocortical Amyloid-beta Burden: A pathways: towards deciphering Alzheimer’s Disease pathogenesis. Candidate Study Using SOMAscan Technology. J Alzheimers Dis Methods Mol Biol 2016;1303:423–432. 2015;46:947–961. [39] Ogishima S, Mizuno S, Kikuchi M, Miyashita A, Kuwano R, [18] Westwood S, Liu B, Baird AL, Anand S, Nevado-Holgado AJ, Tanaka H, et al. A map of Alzheimer’s disease-signaling pathways: Newby D, et al. The influence of insulin resistance on cerebrospinal a hope for drug target discovery. Clin Pharmacol Ther 2013; fluid and plasma biomarkers of Alzheimer’s pathology. Alzheimers 93:399–401. Res Ther 2017;9:31. [40] Hartmann D, Tournoy J, Saftig P, Annaert W, De Strooper B. Implica- [19] Shi L, Baird AL, Westwood S, Hye A, Dobson R, Thambisetty M, et al. tion of APP secretases in notch signaling. J Mol Neurosci 2001; A decade of blood biomarkers for Alzheimer’s Disease Research: an 17:171–181. evolving field, improving study designs, and the challenge of replica- [41] Kopan R, Goate A. Aph-2/Nicastrin: an essential component of tion. J Alzheimers Dis 2018;62:1181–1198. gamma-secretase and regulator of Notch signaling and Presenilin [20] Nakamura A, Kaneko N, Villemagne VL, Kato T, Doecke J, Dore V, localization. Neuron 2002;33:321–324. et al. High performance plasma amyloid-beta biomarkers for Alz- [42] Bosco D, Fava A, Plastino M, Montalcini T, Pujia A. Possible implica- heimer’s disease. Nature 2018;554:249–254. tions of insulin resistance and glucose metabolism in Alzheimer’s dis- [21] Ovod V, Ramsey KN, Mawuenyega KG, Bollinger JG, Hicks T, ease pathogenesis. J Cell Mol Med 2011;15:1807–1821. Schneider T, et al. Amyloid beta concentrations and stable isotope la- [43] Zhao W-Q, Townsend M. Insulin resistance and amyloidogenesis as beling kinetics of human plasma specific to central nervous system common molecular foundation for type 2 diabetes and Alzheimer’s amyloidosis. Alzheimers Dement 2017;13:841–849. disease. Biochim Biophys Acta 2009;1792:482–496. 15525279, 2019, 11, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1016/j.jalz.2019.06.4951 by Readcube (Labtiva Inc.), Wiley Online Library on [20/12/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 1488 L. Shi et al. / Alzheimer’s & Dementia 15 (2019) 1478-1488 [44] Jack CR Jr, Wiste HJ, Weigand SD, Rocca WA, Knopman DS, tau accumulation in amyloid-positive older individuals. Nat Neurosci Mielke MM, et al. Age-specific population frequencies of cerebral 2018;21:424–431. b-amyloidosis and neurodegeneration among people with normal [54] Khan UA, Liu L, Provenzano FA, Berman DE, Profaci CP, Sloan R, cognitive function aged 50–89 years: a cross-sectional study. Lancet et al. Molecular drivers and cortical spread of lateral entorhinal cortex Neurol 2014;13:997–1005. dysfunction in preclinical Alzheimer’s disease. Nat Neurosci 2014; [45] Rowe CC, Ellis KA, Rimajova M, Bourgeat P, Pike KE, Jones G, et al. 17:304–311. Amyloid imaging results from the Australian Imaging, Biomarkers [55] Pooler AM, Polydoro M, Maury EA, Nicholls SB, Reddy SM, and Lifestyle (AIBL) study of aging. Neurobiol Aging 2010; Wegmann S, et al. Amyloid accelerates tau propagation and toxicity 31:1275–1283. in a model of early Alzheimer’s disease. Acta Neuropathol Commun [46] Morris JC, Roe CM, Xiong C, Fagan AM, Goate AM, 2015;3:14. Holtzman DM, et al. APOE predicts amyloid-beta but not tau [56] Gentil BJ, Tibshirani M, Durham HD. Neurofilament dynamics and Alzheimer pathology in cognitively normal aging. Ann Neurol involvement in neurological disorders. Cell Tissue Res 2015; 2010;67:122–131. 360:609–620. [47] Jack CR Jr, Wiste HJ, Weigand SD, Knopman DS, Vemuri P, [57] Lista S, Hampel H. Synaptic degeneration and neurogranin in the path- Mielke MM, et al. Age, sex, and APOE epsilon4 effects on memory, ophysiology of Alzheimer’s disease. Expert Rev Neurother 2017; brain structure, and beta-amyloid across the adult life span. JAMA 17:47–57. Neurol 2015;72:511–519. [58] Baldacci F, Lista S, Cavedo E, Bonuccelli U, Hampel H. Diagnostic [48] Jansen WJ, Ossenkoppele R, Knol DL, Tijms BM, Scheltens P, function of the neuroinflammatory biomarker YKL-40 in Alzheimer’s Verhey FR, et al. Prevalence of cerebral amyloid pathology in persons disease and other neurodegenerative diseases. Expert Rev Proteomics without dementia: a meta-analysis. Jama 2015;313:1924–1938. 2017;14:285–299. [49] Singh PP, Singh M, Mastana SS. APOE distribution in world popula- [59] Querol-Vilaseca M, Colom-Cadena M, Pegueroles J, San Martin- tions with new data from India and the UK. Ann Hum Biol 2006; Paniello C, Clarimon J, Belbin O, et al. YKL-40 (Chitinase 3-like I) 33:279–308. is expressed in a subset of astrocytes in Alzheimer’s disease and other [50] Ashton NJ, Nevado-Holgado AJ, Barber IS, Lynham S, Gupta V, tauopathies. J Neuroinflammation 2017;14:118. Chatterjee P, et al. A plasma protein classifier for predicting amyloid [60] Morris M, Knudsen GM, Maeda S, Trinidad JC, Ioanoviciu A, burden for preclinical Alzheimer’s disease. Sci Adv 2019;5. eaau7220. Burlingame AL, et al. Tau post-translational modifications in wild- [51] McGee S. Simplifying likelihood ratios. J Gen Intern Med 2002; type and human amyloid precursor protein transgenic mice. Nat Neu- 17:646–649. rosci 2015;18:1183–1189. [52] Ciarmiello A, Tartaglione A, Giovannini E, Riondato M, [61] Seward ME, Swanson E, Norambuena A, Reimann A, Cochran JN, Giovacchini G, Ferrando O, et al. Amyloid burden identifies neuropsy- Li R, et al. Amyloid-beta signals through tau to drive ectopic neuronal chological phenotypes at increased risk of progression to Alzheimer’s cell cycle re-entry in Alzheimer’s disease. J Cell Sci 2013; disease in mild cognitive impairment patients. Eur J Nucl Med Mol 126:1278–1286. Imaging 2019;46:288–296. [62] Ittner LM, Ke YD, Delerue F, Bi M, Gladbach A, van Eersel J, et al. [53] Jacobs HIL, Hedden T, Schultz AP, Sepulcre J, Perea RD, Dendritic function of tau mediates amyloid-beta toxicity in Alz- Amariglio RE, et al. Structural tract alterations predict downstream heimer’s disease mouse models. Cell 2010;142:387–397.