15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 Received: 24 July 2020 Revised: 26 November 2020 Accepted: 5 February 2021 DOI: 10.1002/alz.12322 RESEARCH ARTICLE Replication study of plasma proteins relating to Alzheimer’s pathology Liu Shi1 Laura M. Winchester1 Sarah Westwood1 Alison L. Baird1 Sneha N. Anand1 Noel J Buckley1 Abdul Hye2 Nicholas J. Ashton2,3,4 Isabelle Bos5,6 Stephanie J. B. Vos5 Mara ten Kate6 Philip Scheltens6 Charlotte E. Teunissen7 Rik Vandenberghe8 Silvy Gabel8,9 Karen Meersmans8,9 Sebastiaan Engelborghs10,11 Ellen E. De Roeck10,12 Kristel Sleegers13,14 Giovanni B. Frisoni15,16 Olivier Blin17 Jill C. Richardson18 Régis Bordet19 José L. Molinuevo20,21 Lorena Rami21 Anders Wallin22 Petronella Kettunen22 Magda Tsolaki23 Frans Verhey5 Alberto Lléo24 Isabel Sala25 Julius Popp26,27 Gwendoline Peyratout26 Pablo Martinez-Lage28 Mikel Tainta28 Peter Johannsen29 Yvonne Freund-Levi2,30 Lutz Frölich31 Valerija Dobricic32 Cristina Legido-Quigley33,34 Frederik Barkhof35,36 Ulf Andreasson3,37 Kaj Blennow3,37 Henrik Zetterberg3,37,38,39 Johannes Streffer13,40 Christina M. Lill41,42 Lars Bertram32,43 Pieter Jelle Visser5,6 Hartmuth C. Kolb44 Vaibhav A. Narayan44 Simon Lovestone1,44 Alejo J. Nevado-Holgado1 1 Department of Psychiatry, University of Oxford, Oxford, UK 2 Maurice Wohl Clinical Neuroscience, Institute of Psychiatry, Psychology and Neuroscience, Kings College London, London, UK 3 Department of Psychiatry and Neurochemistry, the Sahlgrenska Academy at the University of Gothenburg, Mölndal, Sweden 4 Wallenberg Centre for Molecular and Translational Medicine, University of Gothenburg, Gothenburg, Sweden 5 Department of Psychiatry and Neuropsychology, School for Mental Health and Neuroscience, Alzheimer Centrum Limburg, Maastricht University, Maastricht, the Netherlands 6 Alzheimer Center, VU University Medical Center, Amsterdam, the Netherlands 7 Neurochemistry lab, Department of Clinical Chemistry, Amsterdam University Medical Centers, Vrije Universiteit, Amsterdam Neuroscience, Amsterdam, the Netherlands 8 University Hospital Leuven, Leuven, Belgium 9 Laboratory for Cognitive Neurology, Department of Neurosciences, KU Leuven, Leuven, Belgium 10 Reference Center for Biological Markers of Dementia (BIODEM), Institute Born-Bunge, University of Antwerp, Antwerp, Belgium 11 Department of Neurology, UZ Brussel and Center for Neurociences (C4N), Vrije Universiteit Brussel, Brussels, Belgium 12 Department of Neurology and Memory Clinic, Hospital Network Antwerp (ZNA) Middelheim and Hoge Beuken, Antwerp, Belgium 13 Complex Genetics Group, VIB Center for Molecular Neurology, VIB, Antwerp, Belgium 14 Institute Born-Bunge, Department of Biomedical Sciences, University of Antwerp, Antwerp, Belgium 15 University of Geneva, Geneva, Switzerland 16 IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli, Brescia, Italy 17 AIX marseille university, INS, Ap-hm, Marseille, France 18 Neurosciences Therapeutic Area, GlaxoSmithKline R&D, Stevenage, UK This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. © 2021 The Authors. Alzheimer’s & Dementia published by Wiley Periodicals LLC on behalf of Alzheimer’s Association 1452 wileyonlinelibrary.com/journal/alz Alzheimer’s Dement. 2021;17:1452–1464. 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 SHI ET AL . 1453 19 Inserm, University of Lille, CHU Lille, Lille, France 20 Alzheimer’s Disease and Other Cognitive Disorders Unit, Hopsital Clínic-IDIBAPS, Barcelona, Spain 21 Barcelona Beta Brain Research Center, Unversitat Pompeu Fabra, Barcelona, Spain 22 Institute of Neuroscience and Physiology, Sahlgrenska Academy at University of Gothenburg, Gothenburg, Sweden 23 1st Department of Neurology, AHEPA University Hospital, Makedonia, Thessaloniki, Greece 24 Hospital de la Santa Creu i Sant Pau, Barcelona, Spain 25 Department of Neurology, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain 26 University Hospital of Lausanne, Lausanne, Switzerland 27 Geriatric Psychiatry, Department of Mental Health and Psychiatry, Geneva University Hospitals, Geneva, Switzerland 28 CITA-Alzheimer Foundation, San Sebastian, Spain 29 Danish Dementia Research Centre, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark 30 Karolinska Institutet Center for Alzheimer Research, Division of Clinical Geriatrics, School of Medical Sciences Örebro University and Department of Neurobiology, Caring Sciences and Society (NVS), Stockholm, Sweden 31 Department of Geriatric Psychiatry, Zentralinstitut für Seelische Gesundheit, University of Heidelberg, Mannheim, Germany 32 Lübeck Interdisciplinary Platform for Genome Analytics, Institutes of Neurogenetics and Cardiogenetics, University of Lübeck, Lübeck, Germany 33 Kings College London, London, UK 34 The Systems Medicine Group, Steno Diabetes Center, Gentofte, Denmark 35 Department of Radiology and Nuclear Medicine, VU University Medical Center, Amsterdam, the Netherland 36 UCL Institutes of Neurology and Healthcare Engineering, London, UK 37 Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden 38 UK Dementia Research Institute at UCL, London, UK 39 Department of Neurodegenerative Disease, UCL Institute of Neurology, London, UK 40 UCB, Braine-l’Alleud, Belgium, formerly Janssen R&D, LLC Beerse, Beerse, Belgium 41 Section for Translational Surgical Oncology and Biobanking, Department of Surgery, University of Lübeck and University Medical Center Schleswig-Holstein, Campus Lübeck, Lübeck, Germany 42 Ageing Epidemiology Research Unit, School of Public Health, Imperial College, London, UK 43 Department of Psychology, University of Oslo, Oslo, Norway 44 Janssen R&D, Beerse, UK Correspondence Liu Shi, Department of Psychiatry, University of Abstract Oxford, Oxford, UK. E-mail: liu.shi@psych.ox.ac.uk Introduction: This study sought to discover and replicate plasma proteomic biomarkers relating to Alzheimer’s disease (AD) including both the “ATN” (amy- Funding information loid/tau/neurodegeneration) diagnostic framework and clinical diagnosis. Innovative Medicines Initiative Joint Undertak- ing, Grant/Award Number: n◦ 115372; Euro- Methods: Plasma proteins from 972 subjects (372 controls, 409 mild cognitive impair- pean Union’s Seventh Framework Programme, ment [MCI], and 191 AD) were measured using both SOMAscan and targeted assays, Grant/Award Number: FP7/2007-2013; Euro- pean Commission within the 5th framework including 4001 and 25 proteins, respectively. program, Grant/Award Numbers: QLRT- Results: Protein co-expression network analysis of SOMAscan data revealed the rela- 2001-2455, #37670; Stichting voor Alzheimer Onderzoek, Grant/Award Numbers: #11020, tion between proteins and “N” varied across different neurodegeneration markers, #13007, #15005; Department of Health of the indicating that the ATN variants are not interchangeable. Using hub proteins, age, and Basque Government, Grant/Award Number: 17.0.1.08.12.0000.2.454.01.41142.001.H; apolipoprotein E ε4 genotype discriminated AD from controls with an area under the University of Antwerp Research Fund; curve (AUC) of 0.81 and MCI convertors from non-convertors with an AUC of 0.74. Vetenskapsrådet, Grant/Award Number: #2018-02532; H2020 European Research Targeted assays replicated the relation of four proteins with the ATN framework and Council, Grant/Award Number: #681712; clinical diagnosis. Swedish State Support for Clinical Research, Grant/Award Number: #ALFGBG-720931; UK Discussion: Our study suggests that blood proteins can predict the presence of AD Dementia Research Institute pathology as measured in the ATN framework as well as clinical diagnosis. KEYWORDS Alzheimer’s disease, ATN framework, biomarker, dementia, network analysis, plasma proteomics, replication 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 1454 SHI ET AL . 1 INTRODUCTION RESEARCH IN CONTEXT Alzheimer’s disease (AD) is characterized by the presence of amy- 1. Systematic review: Plasma proteins are studied as loid beta (Aβ) containing plaques and neurofibrillary tangles com- candidate biomarkers to predict the “ATN” (amy- posed of modified tau protein together with the progressive loss of loid/tau/neurodegeneration) diagnostic framework, synapses and neurons.1 Currently, the best characterized methods for while most such studies include small samples. Fur- measuring amyloid or tau pathology are positron emission tomogra- thermore, few studies compare the difference between phy (PET) imaging and cerebrospinal fluid (CSF) measurement.2,3 The different ATN variants from the perspective of plasma National Institute on Aging and Alzheimer’s Association (NIA-AA) have proteins. proposed a biomarker-based framework for classifying AD based on 2. Interpretation: Our findings offer new insights into biomarkers of amyloid pathology (A), tau pathology (T), and neurode- changes in individual proteins and protein networks generation (N).4 Briefly, “A” is measured by cortical amyloid PET lig- linked to various AD pathology markers as well as the and binding or CSF Aβ42; “T” is measured by CSF phosphorylated tau ATN framework, particularly in preclinical stages of AD. (p-tau) or cortical tau PET ligand binding; and “N” is by CSF total tau Our results also indicate that ATN variants are not inter- (t-tau), 18F-fluorodeoxyglucose (FDG) PET, or brain atrophy on mag- changeable because the relation between proteins and netic resonance imaging (MRI). CSF neurofilament light chain (NfL) and “N” vary across different neurodegeneration markers. neurogranin (Ng) may also be potential “N” markers.5,6 Our study is the largest plasma proteomic study of var- However, many of these measures are challenging because of ious AD pathology markers, particularly the ATN frame- their invasiveness, high cost, and limited availability.7,8 Blood-based work, to our knowledge. biomarkers show promise as a less invasive and potentially cost- 3. Future directions: This work suggests that blood pro- effective option for the detection, classification, and monitoring of teins can predict the presence of Alzheimer’s disease AD pathology. We have been seeking to develop multiplexed assays (AD) pathology as measured in the ATN framework. suitable for large-scale research screening using two approaches. In Those nominated proteins are tractable targets for fur- the first, we have used a range of mass-spectrometry proteomic ther mechanistic studies of AD pathology. approaches to generate a panel of “targeted proteins” to be used on multiplexed platforms; in the second we have used a proprietary near- proteomic wide aptamer capture array (SOMAscan).9–13 Recently, we applied these approaches independently to the EMIF-AD Multimodal MBD study. Plasma was available on 972 subjects comprising 372 Biomarker Discovery study (EMIF-AD MBD), specifically in relation CTL, 409 MCI, and 191 AD. Samples were collected from the con- to the detection of amyloid.14,15 Here, we develop this approach fur- stituent cohorts in EMIF-AD using a range of protocols. General ther, now combining both targeted and untargeted proteomics and clinical and demographic information were available for all subjects using protein co-expression analysis and differential expression anal- (including apolipoprotein E [APOE] ε4 genotype data). In addition, each ysis to investigate changes in networks of proteins as well as individual subject had a measure of brain amyloid load, using either CSF Aβ or proteins relating to the ATN framework (Figure 1). We have two objec- amyloid PET. Furthermore, CSF t-tau and p-tau analysis data were tives: first, to compare plasma protein profiles in a variety of ATN vari- available for more than 90% of the subjects. The classification of ants (using different “N” biomarkers) and second to test the replication the status (abnormal/normal) of amyloid, t-tau, and p-tau has been of blood biomarkers relating to the ATN framework as well as to clinical described previously.16 phenotypes. In addition, the following AD-related phenotypes were also mea- sured for the majority of the subjects (Table 1): (1) CSF NfL, Ng, and YKL-40; (2) MRI measures of hippocampal volume, cortical thickness, 2 METHODS and white matter hyperintensities (WMH); (3) clinical assessments including baseline diagnosis, baseline Mini-Mental State Examination 2.1 Participants: EMIF-AD MBD study (MMSE) score, and MCI conversion.16 The status (abnormal/normal) of NfL and hippocampal volume were determined by the median value of The EMIF-AD MBD study is part of the European Medical Information each marker within the whole data set. Framework for AD (EMIF; http://www.emif.eu/emif-ad-2/); a public– We defined the ATN status for each participant using the above private partnership funded through the Innovative Medicines Initia- measurements. Briefly, we used CSF (or where not available, PET) amy- tive (IMI). The design of the EMIF-AD MBD study has been described loid as “A” and CSF p-tau181 as “T.” For “N,” we used CSF t-tau, NfL, previously16 but in brief, ≈1200 samples from three groups of people and hippocampal volume as biomarkers of N. We then dichotomized (cognitively normal controls [CTL], mild cognitive impairment [MCI], these biomarkers as normal or abnormal and categorized them into and AD) were chosen from pre-existing cohorts with the goal of includ- four groups: no pathology (A–T–N–, referring as “A–TN–”), amyloid ing samples from people with pathology as well as those without. All positive but both T and N negative (A+T–N–, referring as “A+TN–”), participating centers have agreed to share data as part of the EMIF-AD amyloid positive and T/N positive (including A+T–N+, A+T+N–, and 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 SHI ET AL . 1455 F I G U R E 1 Flowchart of study design. A, Measurement and quantification of 4001 proteins using SOMAscan assay; (B) Measurement and quantification of 25 proteins using enzyme-linked immunosorbent assay and Luminex. Twenty proteins overlapped between approach (A) and (B) TA B L E 1 Demographics of EMIF participants included in the analysis by diagnosis. Percentage of cases is shown in brackets for male sex, APOE ε4 carriers and the abnormality of amyloid, p-tau, and t-tau Characteristics Sample size CTL MCI AD P N 972 372 409 191 NA Age mean (SD), y 972 64.6 (8.0) 69.9 (8.0) 70.5 (8.8) <.001 Male sex N (%) 972 209 (56) 216 (53) 103 (54) .64 APOE ε4+ N (%) 972 139 (37) 195 (48) 116 (61) <.001 MMSE (SD) 967 28.8 (1.2) 26.2 (2.6) 21.4 (4.7) <.001 Education mean (SD), y 972 12.8 (3.7) 11.0 (3.7) 10.3 (3.9) <.001 Amyloid + N (%) 972 112 (30) 254 (62) 168 (88) <.001 P-tau + N (%) 876 53 (19) 215 (53) 128 (67) <.001 T-tau + N (%) 880 54 (19) 235 (58) 152 (80) <.001 CSF NfL (SD) (pg/mL) 643 742.1 (486.6) 1231.9 (2309.0) 1777.6 (2843.1) <.001 CSF Ng (SD) (pg/mL) 598 125.8 (203.9) 151.3 (199.1) 149.4 (125.3) .33 CSF YKL-40 (SD) (ng/mL) 649 141.9 (57.0) 176.9 (62.1) 191.4 (69.6) <.001 Hippocampal volume (SD) in mm3 633 7628.6 (883.7) 6780.2 (1233.9) 6220.2 (940.1) <.001 Cortical thickness (SD) in mm 586 2.3 (0.1) 2.3 (0.1) 2.3 (0.1) .36 WMH (SD) in mL 617 0.9 (0.7) 1.1 (0.8) 1.0 (0.9) .02 Abbreviations: +, abnormality; AD, Alzheimer’s disease; APOE, apolipoprotein E; CSF, cerebrospinal fluid; CTL, cognitively normal controls; MCI, mild cogni- tive impairment; MMSE, Mini-Mental State Examination; NfL, neurofilament light chain; Ng, neurogranin; p-tau, phosphorylated tau; SD, standard deviation; t-tau, total tau; WMH, white matter hyperintensities. A+T+N+, referring as “A+TN+”), and suspected non-Alzheimer’s teins in plasma. SOMAscan is an aptamer-based assay allowing for the pathology (SNAP, including A–T–N+, A–T+N–, and A–T+N+). simultaneous measurement and quantification of, in the version used here, 4001 proteins. The assay uses chemically modified nucleotides to transform a protein signal into a nucleotide signal that can be 2.2 Plasma analyses quantified using relative fluorescence on microarrays.17 The abun- dance of each protein was log-transformed for all subsequent analy- We used two approaches to measure proteins in plasma samples col- ses. Second, we used enzyme-linked immunosorbent assay (ELISA) and lected from pre-existing cohorts as part of EMIF-AD MBD. First, we Luminex xMAP assays to measure 25 proteins in the same subjects, as used the SOMAscan assay platform (SomaLogic Inc.) to measure pro- described previously.15 Overall, 20 proteins overlapped between the 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 1456 SHI ET AL . SOMAScan array and the second approach (ELISA and Luminex xMAP; WebGestalt software (http://www.webgestalt.org/). Briefly, DE pro- Figure 1). teins or proteins within a module were assembled into a “protein list” and all 4001 proteins measured by the SOMAscan assay were used as “background.” This enrichment analysis was performed on the Kyoto 2.3 Statistical analysis Encyclopediat of Genes and Genomes database. All statistical analyses were completed using R (version 3.3.2). To compare baseline cohort characteristics across three different 2.3.4 Machine learning diagnostic groups (CTL, MCI, and AD), we used one-way analysis of variance (ANOVA) and chi-square tests to compare continuous and Machine learning was used to identify optimal multivariate signatures, binary variables, respectively. including both proteins and demographic data (age, sex, and APOE ε4) as input features, to differentiate AD from CTL, and MCIc from MCIs. The classifier consisted of a two-stage approach (feature selec- 2.3.1 Weighted gene correlation network analysis tion and classifiers) as described previously.14 Briefly, Lasso was used (WGCNA) to select the “n” top input features that best differentiated AD diagnos- tic groups. Support vector machine (SVM) classifiers were then built on The R package WGCNA18 was used to construct a co-expression net- top of these “n” features to predict the outcome under 10-fold cross- work from the proteins obtained from the SOMAScan assay. This clus- validation. For each analysis, the two steps (feature selection and clas- tering is based on calculating correlations between paired variables, sifiers) were performed over 100 iterations, where, in each iteration, soft-threshold transforming them with a power function (i.e., corβ ), the algorithm was allowed to use one feature more than in the previ- and using the result as an adjacency matrix between variables. The ous iteration, starting with 1 feature and finishing with 100. Feature final step applies hierarchical clustering to this adjacency matrix. We selection resulted in a list of features ranked by their correlation with applied this algorithm with default parameters, except for the follow- AD diagnosis, of which the top “n” was selected for subsequent anal- ing settings: soft threshold power beta = 4, minimum module size = 10 ysis. The classifier step was performed using the selected features to proteins, merge cut height = 0.2. The resulting nine modules or groups obtain a model to discriminate AD diagnosis. Those features that pro- of co-expressed proteins were used to calculate module eigenproteins. duced best performance were reported. The eigenprotein-based connectivity (kME) was used to represent the strength of a protein’s correlation with other protein module members. Proteins with high intramodular kME in the top 90th percentile within 3 RESULTS a module were considered hub proteins. The correlation between eigenproteins and AD phenotypes was cal- 3.1 Subject demographics culated, the P values were corrected with false discovery rate (FDR), and corrected P values were presented in a heat map. Furthermore, Demographic information of subjects is shown in Table 1. No significant we used Student’s t-test to assess pairwise difference of eigenproteins difference was observed in the distribution of sex. The CTL group was among different ATN framework and AD diagnostic groups as well as younger and had a lower proportion of APOE ε4 carriers compared to between MCI participants who subsequently converted to dementia the MCI and AD groups. Furthermore, the CTL participants had more (MCIc) within 3 years relative to those whose MCI remained stable years of education and higher MMSE scores. In terms of AD pathology (MCIs). markers, the ratio of abnormality of amyloid, p-tau, and t-tau in AD and MCI individuals was, as expected, significantly higher than in controls. Moreover, CSF NfL and YKL-40 were also significantly higher in AD and 2.3.2 Protein differential expression analysis MCI while no difference was observed for CSF Ng across three differ- ent diagnostic groups. The MCI and AD groups had more hippocampal To compare the association of proteins with the ATN framework, we atrophy and more WMH than the CTL group. used logistic regression to compare proteins in different ATN profiles to “no pathology controls” (A–TN–), adjusting for age and APOE ε4 genotype. P values were corrected using FDR and presented in volcano 3.2 Plasma protein co-expression network plots. To compare the replication of overlapping proteins in different analysis reveals modules linked to AD pathology ATN framework and AD diagnostic groups, we used Student’s t-test to markers assess pairwise difference and presented P values in the box plots. We first performed a network-based analysis of the plasma proteome as reported by the SOMAScan assay using WGCNA. We found nine 2.3.3 Pathway enrichment analysis modules (M) of co-expressed proteins and ranked them based on size from largest (M1; n = 1472 proteins) to smallest (M9; n = 11 Differentially expressed (DE) proteins and proteins within differ- proteins; Table S1 in supporting information). Figure 2A shows the ent modules were further nominated for pathway analysis using clustering of these modules’ concordance according to similarities in 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 SHI ET AL . 1457 F I G U R E 2 Protein modules correlating to Alzheimer’s disease (AD) pathology markers. A, Weighted gene correlation network analysis (WGCNA) of the plasma proteome. This algorithm generated nine modules (M) of co-expressed proteins. Modules are clustered in the network dendrogram based on their relatedness. B, Analysis of the association of module with AD pathology markers. * and ** denote significant correlations P < .05 and P < .001 after false discovery rate (FDR) correction respectively. ∼ indicates corrected P value tend to be significant, .05 < corrected P value < .1. “A,” amyloid; “T,” tau; “N,” neurodegeneration; “V,” vascular; “I,” inflammation; +, abnormality; P-tau, phosphorylated tau; T-tau, total tau; NfL, neurofilament light chain; Ng, neurogranin; WMH, white matter hyperintensity expression patterns. We further investigated the biological signifi- 3.3 Correlation of protein networks with the ATN cance of proteins in each module and found that the modules were framework enriched with various pathways after FDR correction (Table S2 in sup- porting information), such as the metabolic pathways (M1 and M4), We first assessed correlations for each module with the ATN frame- cytokine–cytokine receptor interaction (M2 and M3), PI3K–Akt sig- work where “A,” “T,” and “N” were determined by amyloid, CSF naling pathway (M5), transcriptional misregulation in cancer (M6), and p-tau, and t-tau measurement, respectively. We dichotomized the rel- complement and coagulation cascades (M7). evant biomarkers as normal or abnormal and categorized each indi- We then assessed the module correlations to the AD pathology vidual into one of four groups: A–T–N– (no pathology, n = 273), markers (Figure 2B). We used Aβ as “A”; CSF p-tau levels as a biomarker A+TN– (amyloid pathology, n = 115), A+TN+ (AD pathology, n = of tau (“T”); CSF t-tau, NfL, Ng, and structural atrophy on MRI (hip- 383), and A–TN+ (SNAP, n = 89). We then assessed the correlation pocampal volume and cortical thickness) as biomarkers of “N”; WMH of these ATN profiles with each module eigenprotein. We found that volume as a biomarker for vascular disease burden (“V”); and CSF YKL- four modules (M2 blue, M3 brown, M4 yellow, and M8 pink) showed 40 as a biomarker of inflammation (“I”). a significant difference across ATN profiles (Figure 3A-D). They could Overall, four modules (M2 blue, M3 brown, M4 yellow, and M8 be divided into two groups: (1) those influenced by amyloid pathol- pink) had positive correlations with both “A” amyloid and “T” p-tau ogy (A+); for example, M2 blue and M4 yellow modules showed sig- pathology after FDR correction. For “N,” five modules (M2 blue, M3 nificant difference between A–TN– and A+TN– as well as between brown, M4 yellow, M5 green, and M8 pink) had a positive correla- A+TN+ and A–TN+ profiles. (2) Those influenced by tau pathology and tion, while the M9 magenta module had a negative correlation with neurodegeneration (TN+); for example, M2 blue, M3 brown, and M8 CSF t-tau. Of these, three (M2 blue, M4 yellow, and M9 magenta) pink showed significant increase in A+TN+ compared to A+TN– pro- were consistent in their direction of change and reached statistical sig- file (Figure 3A-D). We also checked the module correlations to the ATN nificance for hippocampal volume (Figure 2B). Although none of the framework in CTL only (n = 372). Of the four modules, two (M2 blue modules reached statistical significance with CSF NfL and Ng or cor- and M4 yellow) had a significant increase in individuals with A+TN+ tical thickness, the association between five modules (M2 blue, M3 and A+TN– profiles compared to A–TN– (Figure S1B and C in support- brown, M5 green, M8 pink, and M9 magenta) and NfL tended to be ing information). significant (corrected P values = .07 for all five modules). The same We then used logistic regression to identify differential expressed tendency was also observed between M3 brown and cortical thick- proteins between different ATN profiles and A–TN– (using t-tau as ness (corrected P value = .06; Figure 2B). For “V” and “I,” one and “N”), adjusting for age and APOE ε4 genotype. Comparing amyloid-only five modules were associated with WMH and YKL-40, respectively pathology (A+TN–) to no pathology (A–TN–), we found that 154 pro- (Figure 2B). teins reached statistical significance (P < .05) but none survived FDR 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 1458 SHI ET AL . F I G U R E 3 Protein modules correlate to amyloid/tau/neurodegeneration (ATN) profile. A-D, The correlation of module profiles with the ATN framework, using t-tau as “N.” E, Volcano plot displaying all proteins differentially expressed between A–TN– and A+TN+ individuals. Proteins passing FDR (q < 0.05) are noted in the same color as the module color. F and G, Top 10 hub proteins within M2 blue and M4 yellow module respectively. SNAP, suspected non-Alzheimer’s pathology; M, module; FDR, false discovery rate correction (q < 0.05; Figure S2A in supporting information). Equally, no focused on the two modules (M2 blue and M4 yellow) that showed a proteins passed FDR comparing A–TN+ (SNAP) to no pathology (A– significant difference across ATN profiles in both all samples and CTL TN–; Figure S2B). However, in contrast, 776 survived FDR comparing only. These two modules had 142 hub proteins (115 from M2 blue AD pathology (A+TN+) to no pathology (A–TN–). As demonstrated in and 27 from M4 yellow), of which 141 proteins (99.3%) were DE in the volcano plot of Figure 3E, the majority of DE proteins were in the A+TN+ individuals, further indicating the consistency between pro- M2 blue module (73.1%), followed by M1 turquoise (11.7%), M4 yellow tein co-expression network analysis and proteome wide differential (9.9%), and M3 brown (3.9%) modules, indicating consistency across analysis. Figure 3F and G showed the top 10 hub proteins within the differential expression analysis and co-expression network analysis. M2 blue and M4 yellow module, respectively (Table S4 in supporting Pathway enrichment analysis further showed that DE proteins were information). enriched in nine pathways such as AD pathway (Table S3 in supporting As we had multiple measures that have been used as markers of information). neurodegeneration or “N,” we explored the relationship of each with In protein co-expression networks, hub proteins are likely impor- the protein networks. Of the four modules (M2 blue, M3 brown, M4 tant proteins because they are highly correlated with other proteins in yellow, and M8 pink) significantly increased in A+TN+ when t-tau the module. We sought to determine whether proteins DE in A+TN+ was used as “N” (Figure 3A-D), three and two remained significant individuals were also hub proteins in the co-expression networks. We when NfL and hippocampal volume were used as “N,” respectively 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 SHI ET AL . 1459 F I G U R E 4 Protein module associations with Alzheimer’s disease (AD) diagnosis (A-D), MCI conversion (E-H), and correlations with cognitive status (I-L). CTL, control; MCI, mild cognitive impairment; MCIc, MCI convertors; MCIs, stable MCI or MCI non-converted to AD; MMSE, Mini-Mental State Examination (Figure S3A-C and Figure S3E and F in supporting information). From blue, M3 brown, M4 yellow, and M8 pink modules showed signifi- differential expressed proteins analysis (A+TN+ vs. A–TN–), we found cant increase in AD compared to MCI and controls (Figure 4A-D). The 151 and 974 proteins passed FDR when NfL and hippocampal volume eigenproteins for three modules (M2 blue, M3 brown, M4 yellow) also were used as “N,” respectively (Figure S4A and B in supporting informa- increased between MCI converters (MCIc, n = 103) and MCI non- tion). One hundred thirty-four proteins were overlapping across three converters (MCIs, n = 223) (Figure 4E-G). In contrast, M9 magenta different “N” markers (Figure S4C), which was significantly higher than decreased in MCI converters (Figure 4H). Furthermore, all four expected by chance alone (P < .001), indicating the similarity of pro- modules (M2 blue, M3 brown, M4 yellow, and M8 pink) were negatively tein profile across three “N” markers. Pathway enrichment of DE pro- correlated with the MMSE score (Figure 4I-L). These results are in con- teins revealed no pathways passed FDR correction when using NfL as cordance with ATN correlations as these four modules showed a strong “N.” In contrast, 42 pathways were enriched using hippocampal volume increase in A+TN+ individuals. as “N,” among which 9 were overlapping with those obtained using t- tau as “N” (Table S3). Considering that more samples had CSF t-tau (n = 880) measurement than NfL (n = 643) or MRI (n = 633) measurements, 3.5 Multiprotein classifier of AD diagnosis and we therefore used CSF t-tau as biomarkers of neurodegeneration (“N”) MCI conversion for the following analysis. Having demonstrated the correlation of protein networks with AD clin- ical groups, we then sought to find a minimal signal from the SOMAS- 3.4 Correlation of protein networks with AD can assay data that might serve as a biomarker for clinical AD diag- diagnosis, MCI conversion, and MMSE score nosis. To do this we built machine learning classifiers with 10-fold cross-validation to identify the optimal multivariate signatures that We further investigated the module correlations to clinical measures differentiated between AD and CTL. We first used demographic vari- including AD diagnosis, MCI conversion, and MMSE score. The M2 ables (age, sex, and education) with APOE ε4 genotype as input features 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 1460 SHI ET AL . F I G U R E 5 Receiver operating characteristic (ROC) curves of models distinguishing Alzheimer’s disease (AD) from controls in (A) whole group and (B) APOE ε4 genotype negative group as well as mild cognitive impairment (MCI) conversion in (C) whole group and (D) APOE ε4 genotype negative group. APOE, apolipprotein E; CTL, control; MCIc, MCI converted to AD; MCIs, stable MCI or MCI non-converted to AD to predict AD diagnosis and found that a combination of age and APOE observed in the APOE ε4 negative group using proteins (AUC = 0.73, ε4 achieved the highest predictive value with an AUC (area under the 95% CI [0.68, 0.85]) compared to age alone (AUC = 0.58, 95% CI [0.51, curve) of 0.72 with 95% confidence intervals (CI; 0.69, 0.78; Figure 5A). 0.71]; Figure 5D). Then, we added 142 hub proteins of M2 blue and M4 yellow modules and identified a panel of eight features achieved the highest predictive value with an AUC of 0.81 (95% CI [0.78, 0.85]; Figure 5A). The input 3.6 Replication of several proteins using ELISA features automatically selected by the classifier were APOE ε4 geno- and Luminex xMAP type and seven proteins (Table S5 in supporting information). We fur- ther performed such comparison in the APOE ε4 negative group and We also measured 25 proteins in the same cohort using targeted found that a model containing nine proteins (Table S5) best discrim- approaches including individual assay (ELISA) and multiplexed plat- inated AD from controls with an AUC of 0.82 (95% CI [0.80, 0.89]), forms such as Luminex xMAP (Figure 1B).15 Overall, 20 of these pro- which was higher than age alone (AUC = 0.73, 95% CI [0.66, 0.80]; teins overlapped with the SOMAScan array. Of these, eight proteins Figure 5B). were DE in A–TN– relative to A+TN+. Four were both consistent in Based on the same approach, we developed a model containing direction of change compared to the SOMAScan assay and reached three proteins (Table S5) that best discriminated MCIc from MCIs. statistical significance: ficolin 2 (FCN2) (Figure 6A and E), plasmino- The diagnostic accuracy was higher (AUC = 0.74, 95% CI [0.71, 0.81]) gen activator inhibitor 1 (PAI-1), C-reactive protein (CRP), and vas- compared to those obtained using age and APOE ε4 genotype alone cular cell adhesion protein 1 (sVCAM1; Figure S5A-F in supporting (AUC = 0.60, 95% CI [0.56, 0.68]; Figure 5C). Higher accuracy was also information). In terms of AD diagnosis, two proteins were differently 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 SHI ET AL . 1461 F I G U R E 6 Correlation of ficolin-2 (FCN2) with amyloid/tau/neurodegeneration (ATN), Alzheimer’s disease (AD) diagnosis, mild cognitive impairment (MCI) conversion, and MMSE form SOMAscan (A-D) and targeted approaches (E-H). CTL, control; MCI, mild cognitive impairment; MCIc, MCI convertors; MCIs, stable MCI or MCI non-converted to AD; MMSE, Mini-Mental State Examination; SNAP, suspected non-Alzheimer’s pathology; altered between AD and CTL from both approaches. They were trols with an AUC of 0.81 and MCIc from MCIs with an AUC of 0.74. FCN2 (Figure 6B and F) and PAI-1 (Figure S6A and B in support- From targeted protein analysis, we replicated the relations of four pro- ing information). FCN2 also showed consistent strong increases in teins with the ATN framework and AD clinical diagnosis. MCIc compared to MCIs from both approaches (Figure 6C and G). The increasing recognition that a broad spectrum of pathologies In addition, two proteins showed consistent significant relation with contribute to AD has highlighted the urgent need for biomarkers that MMSE. Of these, FCN2 had a negative correlation with MMSE (Fig- more comprehensively reflect the complex mechanisms underlying this ure 6D and H) and PAI-1 had positive correlations with MMSE disease.19,20 However, current methods of measuring AD pathology (Figure S6C and D). markers are challenging because of their invasiveness, high cost, and limited availability. To address this challenge, we applied an integra- tive proteomics approach to identify plasma markers linked to a variety 4 DISCUSSION of AD pathology markers including “A,” “T,” “N,” “V,” and “I.” We found four modules had positive correlations with both “A” amyloid and “T” p- In this study, we measured plasma proteins from 972 subjects (372 tau181 pathology. At the time of conducting the study, CSF p-tau 181 CTL, 409 MCI, and 191 AD) using both capture array proteomics was the optimal assay for estimation of pathological tau in biofluids. (SOMAscan) and targeted assays (ELISA and Luminex). For SOMAs- The proteins within these four modules were enriched in various path- can data, we performed both proteome-wide differential analysis and ways that have been reported to be associated with AD such as the Ras protein co-expression network analysis to gain insights into changes in signaling pathway,21 MAPK signaling pathwayy22 and JAK-STAT sig- individual proteins as well as networks of proteins relating to the ATN naling pathway,23 further demonstrating the relatedness of these pro- framework. We found consistent results through both approaches. teins with AD. Five modules were found to be associated with the “I” Another notable finding is that the relation between proteins and “N” marker YKL-40, among which three modules (M1 turquoise, M3 brown, varied across different neurodegeneration markers, suggesting that and M7 black) were enriched in inflammatory response pathways such ATN variants are not interchangeable from the perspective of plasma as cytokine–cytokine receptor interaction and complement and coag- protein profiles. When using t-tau as “N,” we found four modules were ulation cascades (Table S2). related to the ATN framework in all subjects, among which two (M2 The discrepancy of the association between modules and different blue and M4 yellow) were also related to the ATN framework in the “N” markers is noteworthy. As shown in Figure 2, six modules were cognitively normal group. Furthermore, these four modules were also significantly related to CSF t-tau, while none of them reached statis- associated with AD clinical diagnosis and MMSE score. Using hub pro- tical significance with CSF NfL or Ng. A distinct difference was also teins along with age and APOE ε4 genotype discriminated AD from con- observed when using t-tau, NfL, and hippocampal volume as “N” for 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 1462 SHI ET AL . the ATN framework from both protein co-expression network analy- bers of individuals, regardless of APOE ε4 genotype, as part of trials sis and differential analysis. For example, of the four modules associ- recruitment to speed the development of therapeutic interventions ated with the ATN framework when using t-tau as “N,” only three and for AD. two remain significant when using NfL and hippocampal volume as “N,” From the targeted protein analysis, we replicated the relationships respectively. From differentially expression analysis, 776 and 974 pro- of four proteins (FCN2, PAI-1, CRP, and sVCAM1) with the ATN frame- teins were DE in A+TN+ when t-tau and hippocampal volume were work and with AD clinical diagnosis. Our initial discovery-phase stud- used as “N” respectively, while only 151 proteins remained significant ies demonstrated that these four proteins have a relationship with when NfL was used. Our results therefore indicate that ATN variants AD and its pathology.34 Furthermore, these four biomarker candi- are not interchangeable from the perspective of plasma protein pro- dates are also biologically relevant to the disease process. For exam- files. This finding is consistent with a recent study showing that dif- ple, FCN2 and mannose-binding lectin (MBL) are both activators of the ferent ATN variants are not interchangeable from the perspective of lectin complement pathway35 and CSF MBL levels have been shown clinical stage.24 Therefore, although different markers could be used as to be reduced in AD.36 PAI-1 is involved in Aβ accumulation processes “N,” their underlying mechanism and association with clinical outcomes and knocking out the PAI-1 gene or adding PAI-1 inhibitors dramati- are different, suggesting care needs to be taken in using these different cally reduces Aβ burden in the brain of APP/PS1 mice.37,38 CRP inhib- markers when applying the ATN classification framework. ited Aβ40 in a Ca2+-independent manner and interacted with aggre- When using t-tau as “N,” we found consistent results through gated Aβ40 on the fibril-forming pathway.39 sVCAM1, cell adhesion proteome-wide differential analysis and protein co-expression net- molecules, serve as signal transducers that influence the progression of work analysis related to the ATN framework. For instance, ≈87% of neuroinflammation.40 Our replication further demonstrates that these the differential expressed proteins in A+TN+ individuals were in three proteins are involved in AD pathology pathways and could be promis- ATN-related modules (M2 blue, M3 brown, and M4 yellow). Further- ing targets for future mechanistic studies. more, more than 99% of the hub proteins within two modules (M2 blue There are three limitations to our study. First, this is an associa- and M4 yellow) were DE proteins in A+TN+ individuals. Of these hub tion study and thus it precludes causal inference. Further analysis such proteins (Table S4), many have been reported being associated with as implementing Mendelian randomization to integrate genomics and AD, such as the apolipoprotein proteins (Apo) B and D,25,26 comple- proteomics could help to find proteins causing AD pathology. Second, ment component C6 and C7,27,28 kallikrein (KLK6 and KLK12),29,30 the EMIF cohort was designed to be typical of participants who had mitochondrial-related proteins (COX42 and ATP5F1B),31,32 and so on. high ratios of amyloid pathology and APOE ε4 carriers. Therefore they As these hub proteins are in M2 blue and M4 yellow modules, which are not necessarily representative of the broader community as would were related to the ATN framework in individuals with no cognitive be found in epidemiologically derived samples. Thus the present results impairment, namely in the pre-clinical stage, it provides implications should not be generalized to community-based populations without that they are tractable targets for further mechanistic studies of AD further investigation. Third, this is a cross-sectional study and longi- pathology, particularly in preclinical stages of AD. tudinal studies are required to determine the role and mechanisms of As expected, four ATN framework related modules were also highly nominated proteins in AD initiation and progression. expressed in samples from people with AD and negatively associ- Despite this, our study is the largest plasma proteomic study of var- ated with MMSE score. Using hub proteins within M2 blue and M4 ious AD pathology markers, particularly the ATN framework, to our yellow modules, we identified a panel of proteins that can identify knowledge. By applying both proteome-wide differential analysis and study participants with AD as well as predict MCI conversion. For protein co-expression network analysis, our findings offer new insights those 14 selected proteins (Table S5), we further compared the rela- into changes in individual proteins and protein networks linked to AD tions between these proteins with APOE ε4 genotype and age. Results pathology markers as well as the ATN framework in poorly understood showed that all of them were significantly DE between APOE ε4 pos- preclinical stages of AD. Those nominated hub proteins are tractable itive and negative groups. Eight of them were significantly associated targets for further mechanistic studies of AD pathology. Our findings with age (Table S6 in supporting information). Comparing different also suggest that the relationships between plasma proteins and “N” diagnostic groups, we found that the majority of the proteins remained are dependent on the choice of neurodegeneration marker, indicating significantly DE between APOE ε4 positive and negative individuals that the ATN variants are not interchangeable. In addition, it confirms within MCI and CTL groups, but not in AD patients (Table S6). This that highly multiplexed assays are able to predict the presence of AD might be due to the small sample size of AD patients (n = 191). These pathology as measured in the ATN framework as well as clinical diagno- signatures, generated from proteins, had considerably more accuracy sis. This can be potentially applied as a prescreen to preselect patients than age and APOE ε4 alone, the two biggest risk factors of AD. Fur- for further selection procedures for clinical trials, thus reducing the thermore, we found that the predictive value of these protein signa- cost incurred to clinical trials by screen failure. tures was retained also in APOE ε4 negative groups. This is impor- tant because the proportion of APOE ε4 carriers in the general pop- ACKNOWLEDGMENTS ulation is low (10% to 20%)33 and therefore relying on APOE ε4 to This research was conducted as part of the EMIF-AD MBD project, enhance recruitment to clinical trials is problematic. The results that which has received support from the Innovative Medicines Initia- we report here raise the possibility of prescreening in large num- tive Joint Undertaking under EMIF grant agreement n◦ 115372, 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 SHI ET AL . 1463 resources of which are composed of financial contribution from the Engelborghs has received unrestricted research grants from Janssen European Union’s Seventh Framework Programme (FP7/2007-2013) Pharmaceutica and ADx Neurosciences and has served on scientific and EFPIA companies’ in-kind contribution. The DESCRIPA study advisory boards of Biogen, Eisai, Novartis, Nutricia/Danone, all unre- was funded by the European Commission within the 5th frame- lated to this study. Charlotte E. Teunissen advises Roche; received work program (QLRT-2001-2455). The EDAR study was funded by research reagents from ADxNeurosciences and Euroimmun; and has the European Commission within the 5th framework program (con- received contract research and grants from AxonNeurosciences, Bio- tract # 37670). The Leuven cohort was funded by the Stichting voor gen, Boehringer, Brainstorm Therapeutics, Celgene, EIP Pharma, Esai, Alzheimer Onderzoek (grant numbers #11020, #13007, and #15005). Janssen prevention center, Roche, Toyama, and Vivoryon. RV is a senior clinical investigator of the Flemish Research Foun- dation (FWO). The San Sebastian GAP study is partially funded by DATA AVAILABILITY STATEMENT the Department of Health of the Basque Government (allocation The datasets generated and analysed during the current study are 17.0.1.08.12.0000.2.454.01.41142.001.H). We acknowledge the con- available from the EMIF-AD Catalogue via submitted research pro- tribution of the personnel of the Genomic Service Facility at the VIB- posals that have to be approved by the data-owners from each parent U Antwerp Center for Molecular Neurology. The research at VIB- cohort. CMN is funded in part by the University of Antwerp Research Fund. HZ is a Wallenberg Scholar supported by grants from the Swedish REFERENCES Research Council (#2018-02532); the European Research Council 1. Lane CA, Hardy J, Schott JM. Alzheimer’s disease. Eur J Neurol. (#681712); Swedish State Support for Clinical Research (#ALFGBG- 2018;25:59-70. 720931); the Alzheimer Drug Discovery Foundation (ADDF), USA 2. Schöll M, Maass A, Mattsson N, et al. Biomarkers for tau pathology. Mol (#201809-2016862); and the UK Dementia Research Institute at Cell Neurosci. 2019;97:18-33. 3. Cohen AD, Landau SM, Snitz BE, Klunk WE, Blennow K, Zetterberg UCL. FB is supported by the NIHR biomedical research center H. Fluid and PET biomarkers for amyloid pathology in Alzheimer’s dis- at UCLH. ease. Mol Cell Neurosci. 2018;97:3-17. 4. Jack CR, Jr., Bennett DA, Blennow K, et al. NIA-AA research frame- ETHICS STATEMENT work: toward a biological definition of Alzheimer’s disease. Alzheimers Dement. 2018;14:535-562. Written informed consent was obtained from all participants before 5. Zetterberg H, Skillback T, Mattsson N, et al. Association of cere- inclusion in the study. The medical ethics committee at each site brospinal fluid neurofilament light concentration with Alzheimer dis- approved the study. ease progression. JAMA Neurol. 2016;73:60-67. 6. Portelius E, Zetterberg H, Skillback T, et al. Cerebrospinal fluid neu- rogranin: relation to cognition and neurodegeneration in Alzheimer’s CONFLICTS OF INTEREST disease. Brain. 2015;138:3373-3385. Simon Lovstone is named as an inventor on biomarker intellectual 7. Lista S, Faltraco F, Prvulovic D, Hampel H. Blood and plasma-based property protected by Proteome Sciences and King’s College London proteomic biomarker research in Alzheimer’s disease. Prog Neurobiol. unrelated to the current study and within the past 5 years has advised 2013;101-102:1-17. for Optum labs, Merck, SomaLogic and been the recipient of funding 8. de Almeida SM, Shumaker SD, LeBlanc SK, et al. Incidence of post-dural puncture headache in research volunteers. Headache. 2011;51:1503- from AstraZeneca and other companies via the IMI funndinng scheme. 1510. Henrik Zetterberg has served on scientific advisory boards for Denali, 9. Hye A, Lynham S, Thambisetty M, et al. Proteome-based plasma Roche Diagnostics, Wave, Samumed, Siemens Healthineers, Pinteon biomarkers for Alzheimer’s disease. Brain. 2006;129:3042-3050. Therapeutics, and CogRx; has given lectures in symposia sponsored by 10. Thambisetty M, Simmons A, Velayudhan L, et al. Association of plasma clusterin concentration with severity, pathology, and progression in Fujirebio, Alzecure, and Biogen; and is a co-founder of Brain Biomarker Alzheimer disease. Arch Gen Psychiatry. 2010;67:739-748. Solutions in Gothenburg AB (BBS), which is a part of the GU Ven- 11. Kiddle SJ, Sattlecker M, Proitsi P, et al. Candidate blood proteome tures Incubator Program (all unrelated to this study). Jose L. Molin- markers of Alzheimer’s disease onset and progression: a systematic uevo has served on scientific advisory boards for Eli Lilly, Roche Diag- review and replication study. J Alzheimers Dis. 2014;38:515-531. 12. Sattlecker M, Kiddle SJ, Newhouse S, et al. Alzheimer’s disease nostics, Roche, Biogen, Lundbeck, Novartis, IBL, Axovant, Oryzon, and biomarker discovery using SOMAscan multiplexed protein technology. Merck (all unrelated to this study). Alberto Lleo has served on sci- Alzheimers Dement. 2014;10:724-734. entific advisory boards of Fujirebio Europe, Eli Lilly, Novartis, Nutri- 13. Baird AL, Westwood S, Lovestone S. Blood-based proteomic biomark- cia, and Otsuka and is the inventor of a patent on synaptic markers ers of Alzheimer’s disease pathology. Front Neurol. 2015;6:236. 14. Shi L, Westwood S, Baird AL, et al. Discovery and validation of plasma in CSF (all unrelated to this study). Lutz Frolich has received research proteomic biomarkers relating to brain amyloid burden by SOMAscan funding, consultancy fees, or speech honoraria from Allergan, Avid-Eli assay. Alzheimer’s Dement. 2019;15:1478-1488. Lilly, Avanir, Avraham Pharmaceuticals, Axon Neuroscience, Axovant, 15. Westwood S, Baird AL, Anand SN, et al. Validation of plasma proteomic Biogen, Boehringer Ingelheim, Eisai, Functional Neuromodulation, GE biomarkers relating to brain amyloid burden in the EMIF-Alzheimer’s disease multimodal biomarker discovery cohort. J Alzheimer’s Dis. Health Care, Lundbeck, MerckSharpe&Dohme, Novartis, Pfizer, Pira- 2020:1-13. mal Imaging, Roche, and Schwabe Pharma. Julius Popp has served 16. Bos I, Vos S, Vandenberghe R, et al. The EMIF-AD multimodal on scientific advisory boards of Fujirebio Europe, Eli Lilly, and Nestlé biomarker discovery study: design, methods and cohort characteris- Institute of Health Sciences, all unrelated to this study. Sebastiaan tics. Alzheimers Res Ther. 2018;10:64. 15525279, 2021, 9, Downloaded from https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.12322 by Readcube (Labtiva Inc.), Wiley Online Library on [14/02/2023]. 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 1464 SHI ET AL . 17. Gold L, Ayers D, Bertino J, et al. Aptamer-based multiplexed proteomic 32. Goldberg J, Currais A, Prior M, et al. The mitochondrial ATP syn- technology for biomarker discovery. PLoS One 2010;5:e15004. thase is a shared drug target for aging and dementia. Aging Cell. 18. Langfelder P, Horvath S. WGCNA: an R package for weighted correla- 2018;17:e12715. tion network analysis. BMC Bioinformatics 2008;9:559. 33. Singh PP, Singh M, Mastana SS. APOE distribution in world popula- 19. Lista S, Zetterberg H, O’Bryant SE, Blennow K, Hampel H. Evolving tions with new data from India and the UK. Ann Hum Biol. 2006;33:279- relevance of neuroproteomics in Alzheimer’s disease. Neuroproteomics. 308. Springer; 2017:101-115. 34. Shi L, Baird AL, Westwood S, et al. A decade of blood biomark- 20. Castrillo JI, Lista S, Hampel H, Ritchie CW. Systems biology methods ers for Alzheimer’s disease research: an evolving field, improving for Alzheimer’s disease research toward molecular signatures, sub- study designs, and the challenge of replication. J Alzheimers Dis. types, and stages and precision medicine: application in cohort studies 2018;62:1181-1198. and trials. Biomarkers Alzheimer’s Dis. Drug Dev.. Springer; 2018:31-66. 35. Fujita T, Matsushita M, Endo Y. The lectin-complement pathway–its 21. Kirouac L, Rajic AJ, Cribbs DH, Padmanabhan J. Activation of Ras-ERK role in innate immunity and evolution. Immunol Rev. 2004;198:185- signaling and GSK-3 by amyloid precursor protein and amyloid beta 202. facilitates neurodegeneration in Alzheimer’s disease. ENeuro. 2017;4. 36. Lanzrein A-S, Jobst KA, Thiel S, et al. Mannan-binding lectin in human 22. Ogishima S, Mizuno S, Kikuchi M, et al. A map of Alzheimer’s disease- serum, cerebrospinal fluid and brain tissue and its role in Alzheimer’s signaling pathways: a hope for drug target discovery. Clin Pharmacol disease. Neuroreport. 1998;9:1491-1495. Ther. 2013;93:399-401. 37. Liu RM, van Groen T, Katre A, et al. Knockout of plasminogen acti- 23. Nevado-Holgado AJ, Ribe E, Thei L, et al. Genetic and real-world clini- vator inhibitor 1 gene reduces amyloid beta peptide burden in a cal data, combined with empirical validation, nominate jak-stat signal- mouse model of Alzheimer’s disease. Neurobiol Aging. 2011;32:1079- ing as a target for Alzheimer’s disease therapeutic development. Cells. 1089. 2019;8:425. 38. Akhter H, Huang WT, van Groen T, Kuo HC, Miyata T, Liu RM. A 24. Mattsson-Carlgren N, Leuzy A, Janelidze S, et al. The implications of small molecule inhibitor of plasminogen activator inhibitor-1 reduces different approaches to define AT (N) in Alzheimer disease. Neurology. brain amyloid-β load and improves memory in an animal model of 2020;94(21):e2233-e2244. Alzheimer’s disease. J Alzheimers Dis. 2018;64:447-457. 25. Kalman J, McConathy W, Araoz C, Kasa P, Lacko AG. Apolipopro- 39. Ozawa D, Nomura R, Mangione PP, et al. Multifaceted anti- tein D in the aging brain and in Alzheimer’s dementia. Neurol Res. amyloidogenic and pro-amyloidogenic effects of C-reactive protein 2000;22:330-336. and serum amyloid P component in vitro. Sci Rep. 2016;6:29077. 26. Ishii M. Apolipoprotein B as a new link between cholesterol and 40. Wennström M, Nielsen HM. Cell adhesion molecules in Alzheimer’s Alzheimer disease. JAMA Neurol. 2019;76:751-753. disease. Degener Neurol Neuromuscul Dis. 2012;2:65-77. 27. Crehan H, Hardy J, Pocock J. Microglia, Alzheimer’s disease, and com- plement. Int J Alzheimer’s Dis. 2012;2012:983640. 28. Shen Y, Yang L, Li R. What does complement do in Alzheimer’s disease? SUPPORTING INFORMATION Old molecules with new insights. Transl Neurodegener. 2013;2:1-11. Additional supporting information may be found online in the Support- 29. Zamolodchikov D, Chen Z-L, Conti BA, Renné T, Strickland S. Acti- vation of the factor XII-driven contact system in Alzheimer’s dis- ing Information section at the end of the article. ease patient and mouse model plasma. Proc Natl Acad Sci U S A. 2015;112:4068-4073. 30. Goldhardt O, Warnhoff I, Yakushev I, et al. Kallikrein-related pepti- dases 6 and 10 are elevated in cerebrospinal fluid of patients with How to cite this article: Shi L, Winchester LM, Westwood S, Alzheimer’s disease and associated with CSF-TAU and FDG-PET. Transl et al. Replication study of plasma proteins relating to Neurodegener. 2019;8:1-13. Alzheimer’s pathology. Alzheimer’s Dement. 31. Cardoso SM, Proença MT, Santos S, Santana I, Oliveira CR. Cytochrome c oxidase is decreased in Alzheimer’s disease platelets. 2021;17:1452–1464. https://doi.org/10.1002/alz.12322 Neurobiol Aging. 2004;25:105-110.