Alzheimer’s & Dementia 15 (2019) 776-787 Featured Article Inflammatory biomarkers in Alzheimer’s disease plasma Angharad R. Morgana, Samuel Toucharda, Claire Leckeya, Caroline O’Hagana, Alejo J. Nevado-Holgadob, NIMA Consortiumy, Frederik Barkhofc,d, Lars Bertrame, Olivier Blinf, Isabelle Bosg, Valerija Dobricich, Sebastiaan Engelborghsi,j, Giovanni Frisonik,l, Lutz Fr€ olichm, n o p q Silvey Gabel , Peter Johannsen , Petronella Kettunen , Iwona K1oszewska , Cristina Legido-Quigleyd,r, Alberto Lleos, Pablo Martinez-Laget, Patrizia Mecocciu, Karen Meersmansn, Jose Luis Molinuevov, Gwendoline Peyratoutw, Julius Poppx, Jill Richardsony, Isabel Salaz, Philip Scheltensaa, Johannes Strefferbb, Hikka Soininencc, Mikel Tainta-Cuezvadd, Charlotte Teunissenee, Magda Tsolakiff, Rik Vandenberghegg, Pieter Jelle Visserhh, Stephanie Vosg, Lars-Olof Wahlundii, Anders Wallinjj, Sarah Westwoodb, Henrik Zetterbergkk,ll,mm,nn, Simon Lovestoneb, B. Paul Morgana,*, Annex: NIMA–Wellcome Trust Consortium for Neuroimmunology of Mood Disorders and Alzheimer’s Disease a Systems Immunity Research Institute and UK Dementia Research Institute Cardiff, School of Medicine, Cardiff University, Cardiff, UK b Department of Psychiatry, University of Oxford, Oxford, UK c Department of Radiology and Nuclear Medicine, VU University Medical, Amsterdam, the Netherlands d UCL Institutes of Neurology and Healthcare Engineering, University College London, London, UK e Max Planck Institute for Molecular Genetics, Berlin, Germany f Aix-Marseille University, APHM, Institute Neurosci System, Pharmacology, Marseille, France g Alzheimer Centrum Limburg, Maastricht University, Maastricht, the Netherlands h L€ubeck Interdisciplinary Platform for Genome Analytics, University of L€ubeck, L€ubeck, Germany i Department of Neurology, Hospital Network Antwerp (ZNA), Antwerp, Belgium j Reference Center for Biological Markers of Dementia, Institute Born-Bunge, Antwerp, Belgium k University of Geneva, Geneva, Switzerland l IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli, Brescia, Italy m Department of Geriatric Psychiatry, Zentralinstitut f€ur Seelische Gesundheit, University of Heidelberg, Mannheim, Germany n Department of Neurosciences, Laboratory for Cognitive Neurology, KU Leuven, Leuven, Belgium o Division of Clinical Geriatrics, Department of Neurobiology, Caring Sciences and Society, Karolinska Institutet, Stockholm, Sweden p University of Gothenburg, Institute of Neuroscience and Physiology, Gothenburg, Sweden q Department of Old Age Psychiatry & Psychotic Disorders, Medical University of Lodz, Lodz, Poland r School of Public Health, Imperial College London, London, UK s Department of Neurology, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain t CITA-Alzheimer Foundation, San Sebastian, Spain B.P.M. is a consultant for GlaxoSmithKline (GSK), Roche, Alexion, and Institute of Health Sciences. J.R. was a full-time employee of GSK and is Achillion. F.B. is supported by the NIHR UCLH biomedical research center. a GSK share holder. P.S. has acquired grant support (for the institution) S.E. reports research funding from Janssen Pharmaceutica N.V. and ADx from Piramal. In the past 2 years, he has received consultancy/speaker Neurosciences (paid to institution). L.F. has received honoraria for consul- fees (paid to the institution) from Biogen and Roche (Diagnostics). He is ting or educational lectures, and/or advisory boards from Allergan, Eli Lilly, PI of studies with Probiodrug and EIP Pharma. S.V. receives research sup- Avraham Pharma, Axon Neuroscience, Axovant, Biogen, Boehringer Ingel- port from the Memorabel program of ZonMw (the Netherlands Organiza- heim, Eisai, Functional Neuromodulation, H. Lundbeck A/S, Merck Sharpe tion for Health Research and Development), Janssen Pharmaceutica N.V., & Dohme, Novartis, Pfizer, Piramal Imaging, Pharmatropix, Pharnext, and the Alzheimer’s Association. y Roche, and Willmar Schwaber. P.M.-L. has received honoraria for lecturing See Annex. from Lilly, Nutricia, Stada, Schwabbe, General-Electric, and for partici- *Corresponding author. Tel.: 144 (0)29 2068 7096. pating in advisory boards from Biogen, Nutricia, Lilly, General-Electric. E-mail address: morganbp@cardiff.ac.uk J.P. received grants from the Swiss National Science Foundation (SNF 320030L_141179), Fujirebio Europe, Lilly, Ono Pharma, and the Nestle https://doi.org/10.1016/j.jalz.2019.03.007 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/). A.R. Morgan et al. / Alzheimer’s & Dementia 15 (2019) 776-787 777 u Department of Medicine, Institute of Gerontology and Geriatrics, University of Perugia, Perugia, Italy v Barcelona Beta Brain Research Center, Unversitat Pompeu Fabra, Barcelona, Spain w Department of Psychiatry, Old Age Psychiatry, Lausanne University Hospital, Lausanne, Switzerland x Hopitaux Universitaires Geneve and Universite de Geneve, Geneva, Switzerland y Neurosciences Therapeutic Area, GlaxoSmithKline R&D, Stevenage, UK z Memory Unit, Neurology Department, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain aa Alzheimer Center, Amsterdam University Medical Centers, Vrije Universiteit, Amsterdam, the Netherlands bb Reference Center for Biological Markers of Dementia (BIODEM), Institute Born-Bunge, University of Antwerp, Antwerp, Belgium cc Institute of Clinical Medicine, Neurology, University of Eastern Finland, Kuopio, Finland dd Center for Research and Advanced Therapies. CITA-Alzheimer Foundation, San Sebastian, Spain ee University Hospital Leuven, Leuven, Belgium ff 1st Department of Neurology, AHEPA University Hospital, Makedonia, Thessaloniki, Greece gg Department of Clinical Chemistry, Neurochemistry lab, Amsterdam University Medical Centers, Amsterdam, the Netherlands hh Department of Psychiatry & Neuropsychology, School for Mental Health and Neuroscience, Maastricht University, Maastricht, the Netherlands ii NVS-Department, Section of Clinical Geriatrics, Karolinska Institutet, Huddinge, Sweden jj Section for Psychiatry and Neurochemistry, Institute of Neuroscience and Physiology, University of Gothenburg Sahlgrenska Academy, Gothenburg, Sweden kk Clinical Neurochemistry Lab, Institute of Neuroscience and Physiology, Sahlgrenska University Hospital, M€olndal, Sweden ll Institute of Neuroscience and Physiology, Department of Psychiatry and Neurochemistry, University of Gothenburg, M€olndal, Sweden mm Department of Molecular Neuroscience, UCL Institute of Neurology, London, UK nn UK Dementia Research Institute, London, UK Abstract Introduction: Plasma biomarkers for Alzheimer’s disease (AD) diagnosis/stratification are a “Holy Grail” of AD research and intensively sought; however, there are no well-established plasma markers. Methods: A hypothesis-led plasma biomarker search was conducted in the context of international multicenter studies. The discovery phase measured 53 inflammatory proteins in elderly control (CTL; 259), mild cognitive impairment (MCI; 199), and AD (262) subjects from AddNeuroMed. Results: Ten analytes showed significant intergroup differences. Logistic regression identified five (FB, FH, sCR1, MCP-1, eotaxin-1) that, age/APOε4 adjusted, optimally differentiated AD and CTL (AUC: 0.79), and three (sCR1, MCP-1, eotaxin-1) that optimally differentiated AD and MCI (AUC: 0.74). These models replicated in an independent cohort (EMIF; AUC 0.81 and 0.67). Two analytes (FB, FH) plus age predicted MCI progression to AD (AUC: 0.71). Discussion: Plasma markers of inflammation and complement dysregulation support diagnosis and outcome prediction in AD and MCI. Further replication is needed before clinical translation. 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: Alzheimer’s disease; Biomarker; Plasma; Inflammation; Complement 1. Introduction and practicability of obtaining peripheral blood to measure disease biomarkers make this an attractive option for early Alzheimer’s disease (AD) is a complex neurodegenera- diagnosis and large-scale screening. Numerous discovery tive disorder that develops gradually and progressively, studies for blood-based biomarkers of AD have been re- with symptoms progressing over time from mild forgetful- ported, but validation and replication remain key challenges ness to severe mental impairment. Early diagnosis is an and none has yet achieved clinical usefulness [4–7]. essential requirement for effective intervention but is chal- Promising candidates do exist, for example, plasma Ab42/ lenging because of current reliance on clinical observation 40 ratio and neurofilament light chain [8], but more work and cognitive testing, with diagnosis confirmed postmortem is needed. by demonstrating typical AD brain pathology. Biomarkers of Considerable evidence implicates inflammation and early disease might address this challenge and are thus an complement dysregulation in AD pathogenesis. Genome- urgent unmet need. wide association studies demonstrated strong associations Currently, cerebrospinal fluid (CSF) levels of amyloid b between AD and common SNPs in the gene encoding the (Ab) fragments and hyperphosphorylated or total tau are complement regulator clusterin (CLU) [9]. A second the most widely used biomarkers for AD [1,2]; however, genome-wide association study replicated the CLU associ- diagnostic accuracy varies between centers [3]. Further- ation and identified association with an SNP in the CR1 more, lumbar puncture is invasive and difficult to implement gene, encoding complement receptor 1 (CR1) [10]. These in the presymptomatic elderly population. The accessibility findings have been robustly replicated in diverse cohorts. 778 A.R. Morgan et al. / Alzheimer’s & Dementia 15 (2019) 776-787 Furthermore, pathway analysis has highlighted immunity, from all individuals in the cohorts and stocks had been ex- inflammation, and complement as key pathways in AD hausted for others. Diagnostic categories were created using [11–13]. Other evidence implicating inflammation and similar algorithms in the discovery and replication cohorts complement includes longitudinal studies demonstrating [19–22]. In all cohorts, the definition for CTL was a that inflammation occurs years before AD onset [14,15], normal performance on neuropsychological assessment and cross-sectional studies reporting increased inflamma- (within 1.5 SD of the average for age, gender, and tory markers in early AD [16]. Plasma markers of inflamma- education). Diagnosis of MCI was made according to the tion and complement dysregulation may therefore be useful criteria of Petersen [23], and AD-type dementia was diag- biomarkers of early AD. Indeed, complement proteins, reg- nosed using the National Institute of Neurological and ulators, and activation products were altered in AD plasma Communicative Disorders and Stroke–Alzheimer’s Disease and/or CSF [17], and in a systematic review of 21 discovery and Related Disorders Association criteria [24]. or panel-based blood proteomic studies, complement was Patient data available differed between the cohorts; there- the top implicated pathway across the studies [18]. fore, a minimal clinical data set was collected and harmo- The underpinning hypothesis of this study is that plasma nized as described [21]; this data set comprised 1) levels of complement proteins and other inflammatory bio- demographics: age, gender, education; 2) clinical informa- markers differ between neurologically normal elderly con- tion: diagnosis, medication use, comorbidities, family trols (CTL) and those with mild cognitive impairment history of dementia, functional impairment rating; 3) cogni- (MCI) and/or AD, between subjects with MCI and those tive data: Mini–Mental State Examination, neuropsycholog- with AD, and between subjects with MCI destined to rapidly ical testing. Imaging data and CSF samples were not progress to AD (progressors) and those who will not prog- available for a majority of cases included in the cohorts ress (nonprogressors). If proven, then the most informative and so could not be included in the analyses; however, this of these plasma biomarkers can be used to diagnose, stratify, was not considered an issue given that the aim of the work predict disease progression, and/or demonstrate response to was to identify plasma markers that correlated with clinical intervention in MCI and AD. Analytes were selected based disease status. on biological evidence and published studies of inflamma- tory/complement biomarkers in neurodegeneration. In the 2.2. Discovery phase assays discovery phase, we used singleplex and multiplex solid- phase enzyme immunoassays to measure 53 proteins In the discovery phase, 53 plasma analytes were measured comprising complement components, activation products using commercial and in-house singleplex and multiplex as- and regulators, cytokines and chemokines in a large cohort says on all available samples in duplicate from AddNeur- comprising AD, MCI, and CTL samples. Proteins demon- oMed. Plasma clusterin, soluble complement receptor 2, strating association with AD and/or MCI in this discovery C-reactive protein (CRP), colony-stimulating factor 1 sample set were investigated further in two independent (CSF1), and interleukin-23 (IL-23) were determined using replication cohorts. commercially available enzyme-linked immunosorbent as- says (clusterin, CRP, CSF1, and IL-23 from R&D systems (Abingdon, UK; cat# DY5874, DY1707, DY216, and 2. Methods DY5265 B) and soluble complement receptor 2 from Sino Biological (Beijing, China; cat# SEKA10811); protocols 2.1. Study population were as recommended by the manufacturers. Plasma soluble Discovery phase samples were from AddNeuroMed, a complement receptor 1 (sCR1), C1-inhibitor (C1inh), C5, cross-European cohort for biomarker discovery, detailed C9, C1q, factor H-related protein 4 (FHR4), factor H (FH) elsewhere [19,20]. Informed consent was obtained Y402, and H402 variants were determined using optimized according to the Declaration of Helsinki (1991), and antibody pairs in in-house enzyme-linked immunosorbent as- protocols and procedures were approved by Institutional says as described [25]. Ten complement biomarkers were Review Boards at each collection site. We used 720 measured using customized V-plex electrochemilumines- plasma samples from the cohort: 262 AD, 199 MCI, and cence (ECL) immunoassays (MSD; Rockville, Maryland); 259 CTL, selected based solely on availability of plasma antibody pairs were developed and optimized in-house. Multi- samples. The replication cohorts comprised (1) 867 plasma plex 1 comprised abundant analytes C3, C4, factor B (FB), FH, samples (88 AD, 425 MCI, 347 CTL) from European and factor I (FI). Multiplex 2 comprised low-concentration an- Medical Information Framework for Alzheimer’s Disease alytes factor D (FD); the activation fragments Bb, C3a, and Multimodal Biomarker Discovery (EMIF-AD MBD), a iC3b; and the terminal complement complex (TCC). A cali- cross-European biomarker discovery cohort [21]; (2) 427 bration curve comprising five-fold dilutions of a mixture of plasma samples (105 AD, 69 MCI, 253 CTL) from Mauds- protein standards was run in duplicate on each plate. ECL ley Biomedical Research Centre Dementia Case Registry signal was measured on the MESO QuickPlex SQ 120 reader (DCR) [22]. In both cases, samples were selected based (MSD). Data acquisition and analysis was performed using solely on availability of plasma; plasma was not collected MSD software Discovery workbench 4.0. A.R. Morgan et al. / Alzheimer’s & Dementia 15 (2019) 776-787 779 The V-Plex Human Cytokine 30-Plex Kit (MSD; cat# performed using the Dunn test with Bonferroni correction. K15054D) was used to measure cytokines/chemokines. For 12 analytes (eotaxin-3, granulocyte-macrophage col- The kit comprises three 10-plex panels: V-plex Proinflam- ony-stimulating factor, IL-1b, IL-2, IL-4, IL-5, IL-7, IL-10, matory Panel 1 measures interferon g, interleukin (IL)-1b, IL12p70, IL-13, MIP-1a, TNFb), many samples were below IL-2, IL-4, IL-6, IL-8, IL-10, IL-12p70, IL-13, and tumor assay detection limits; these were analyzed as binary variables necrosis factor (TNF)-a in samples diluted 1:2 in proprietary (positive or negative) and tested for association with disease buffer; V-plex cytokine panel 1 measures granulocyte- status by chi-square test. macrophage colony-stimulating factor, IL-1a, IL-5, IL-7, IL-12/IL-23p40, IL-15, IL-16, IL-17A, TNF-b, and vascular 2.4.2. Identification of optimal analyte sets endothelial growth factor–A in samples diluted 1:4; V- Stepwise logistic regression (SLR) was used to find the plex chemokine panel 1 measures eotaxin-1, macrophage analyte set that optimally distinguished between diagnostic inflammatory protein (MIP)-1b, eotaxin-3, thymus- and groups: CTL versus AD, CTL versus MCI, MCI versus activation-regulated chemokine (TARC; CCL17), inter- AD. Demographic covariates age, gender, and apolipopro- feron-g-inducible protein (IP)-10, MIP-1a, IL-8, MCP-1, tein E (APOE) genotype were controlled for and included macrophage-derived chemokine, and MCP-4 in samples in models as potential predictors. For each comparison, diluted 1:4. All assays were performed according to manu- the data set was randomly split into training (80%) and facturer’s instructions using ECL detection as mentioned validation (20%) sets. The training set was used to select previously. Intra-assay and interassay limits for coefficients variables and fit the model which was then tested on the of variation (CV) were set at 25%, and data for samples with validation set using receiver operating curve (ROC) anal- a CV above this were not included in the analysis. ysis. The models developed for AD versus CTL and MCI versus AD were tested in the replication cohorts using 2.3. Replication phase assays ROC analysis. The analytes selected from the discovery phase for repli- 2.4.3. Markers of disease progression cation were sCR1, FB, FH, MCP-1, and eotaxin-1; FI, TCC, Data on MCI progression to AD were only available in a clusterin, and C4 were also included to replicate previously subset of the EMIF-AD MBD; in this case, SLR was used to reported association of these biomarkers with MCI progres- find the analyte set that best distinguished individuals who sion [26], not tested in AddNeuroMed cohort. subsequently progressed from MCI to AD from nonprogres- MSD U-PLEX custom multiplexing was used in replica- sors. Because the MCI conversion group was relatively tion phase to build bespoke panels. Plasma levels of low small, stepwise selection was performed on the complete abundance analytes sCR1, MCP-1, eotaxin-1, and TCC data set, followed by ROC analysis with leave-one-out were measured in one panel with samples diluted 1:2; cross-validation. To avoid overfitting, 500 replications of plasma levels of high abundance analytes FB, FH, FI, clus- stepwise models were performed on random data subsets, terin, and C4 were measured in a second panel with samples each comprising a training set (80%) for selection and a vali- diluted 1:2000. Assays were performed according to the dation set (20%) for model testing, and ROC analysis per- manufacturer’s instructions using ECL detection. Both formed for each replication. The variables most often panels were run on all EMIF-AD MBD and DCR samples selected and significant were identified. in duplicate. Intra-assay and interassay limits for CV were set at 25% as above. 2.4. Statistical analysis 3. Results All statistical tests and analyses were performed with R 3.1. Individual analytes differ between discovery set software, including ggplot2, caret, and pROC packages. In groups all cases, P , .05 was considered statistically significant. Of the 53 plasma proteins measured in the discovery set, 2.4.1. Individual analytes 10 demonstrated significant differences between clinical Protein concentrations were determined automatically groups (Table 1). Pairwise comparisons (Dunn test with from standard curves plotted using GraphPad Prism5. Values Bonferroni correction) showed (1) for AD versus CTL, were adjusted for recruitment center and plasma storage time increased C4 and eotaxin-1, decreased sCR1, C5, and as described [27] using a generalized linear regression model. CRP; (2) for MCI versus CTL, increased FH, C3, and All subsequent analyses were performed on generalized MCP-1, decreased C5 and MIP-1b; (3) for AD versus linear regression model–adjusted data and log-transformed MCI, increased eotaxin-1 and MIP-1b, decreased FI, C3, to achieve normal distribution. In the discovery phase, associ- CRP, MCP-1 (Table 1; Fig. 1). Of the 12 MSD cytokine/che- ation of individual analytes with disease status was tested us- mokine panel analytes analyzed categorically, none showed ing the Kruskal-Wallis test. Pairwise comparisons were then significant differences between clinical groups. 780 A.R. Morgan et al. / Alzheimer’s & Dementia 15 (2019) 776-787 Table 1 Ten analytes associated with clinical state in the discovery phase Mean 6 SD Mean 6 SD Mean 6 SD P value P value P value P value Analyte CTL (n 5 259) MCI (n 5 199) AD (n 5 262) KW test AD vs. CTL AD vs. MCI MCI vs. CTL FH (mg/ml) 241.5 (56.4) 262.7 (71.8) 258.2 (73.0) .01 ns ns .004 FI (mg/ml) 31.5 (7.0) 32.2 (6.9) 31.0 (7.5) .049 ns .03 ns sCR1 (ng/ml) 11.52 (3.03) 11.43 (3.10) 10.88 (3.01) .043 .03 ns ns C3 (mg/ml) 1042.7 (553.4) 1105.0 (377.4) 1004.2 (435.4) ,.0001 ns .0001 .001 C4 (mg/ml) 351.6 (129.6) 370.8 (136.2) 386.1 (159.3) .01 .01 ns ns C5 (mg/ml) 84.9 (16.2) 81.0 (14.7) 79.8 (14.7) .001 .0004 ns .03 CRP (ng/ml) 996.8 (1145.6) 841.3 (711.1) 761.1 (810.5) .007 .01 .09 ns MCP-1 (pg/ml) 63.1 (22.5) 68.5 (24.5) 63.0 (20.4) .009 ns .006 .002 Eotaxin-1 (pg/ml) 141.6 (65.0) 143.3 (66.2) 162.5 (78.7) ,.0001 ,.0001 ,.0001 ns MIP-1b (pg/ml) 58.9 (29.2) 58.1 (55.2) 63.1 (56.2) .007 ns .006 .002 Abbreviations: AD, Alzheimer’s disease; CRP, C-reactive protein; CTL, control; KW, Kruskal-Wallis; MCI, mild cognitive impairment; ns, not significant; SD, standard deviation. NOTE. Ten analytes showed statistically significant differences in concentration between clinical groups in the discovery phase. The table shows means and standard deviations, KW test P value, and Dunn test P values for each analyte. 3.2. Developing models to differentiate groups dent predictive value. Smaller analyte sets were poor predic- tors (details not shown). We concluded that there was no 3.2.1. AD from CTL reliable and practicable biomarker set from the analytes Stepwise selection demonstrated strong interdependence measured that distinguished MCI and CTL. between some analytes and revealed other analytes that significantly and independently contributed to distinguish- 3.2.4. MCI progressors from nonprogressors ing clinical groups. SLR modeling was used to identify the Baseline samples from 285 individuals with MCI who most predictive set of analytes. A model combining FB, had either progressed to AD when reassessed 12 months FH, sCR1, MCP-1, and eotaxin-1 with covariates age and later (progressors; 55) or had remained stable over this APOε4 status best differentiated AD versus CTL. FH and period (nonprogressors; 230) were compared in EMIF-AD eotaxin-1 were higher and FB, CR1, and MCP-1 were lower MBD. Of the nine analytes measured, only two, FB (higher in AD compared to CTL. Diagnostic accuracy in distin- in progressors) and FH (lower in progressors), were signifi- guishing CTL from AD was moderate (AUC 0.79); 77% of cantly different between progressors and nonprogressors. A samples were predicted correctly with 84% sensitivity and model combining these two analytes with age, the only sig- 70% specificity (Fig. 2A; Table 2). This model was tested nificant covariable, was moderately predictive (AUC 0.71); in the replication cohorts. In EMIF-AD MBD, comprising 67% of samples correctly predicted, sensitivity 71%, speci- 867 plasma samples (88 AD, 425 MCI, 347 CTL), the model ficity 67% (Table 3). In the 500 replications of stepwise strongly replicated (AD vs. CTL; AUC 0.81), correctly pre- models, age and FH were always selected and significant dicting 76% of samples with 73% sensitivity and 77% spec- 499 times, FB was selected 414 times and significant 309 ificity. In DCR, comprising 427 samples (105 AD, 69 MCI, times. No other analyte was selected more than 67 times. 253 CTL), the model performed poorly (AD vs. CTL; AUC The average AUC for the 500 replications was 0.69 (SD 0.58). 0.09). 3.2.2. AD from MCI A model combining sCR1, MCP-1, and eotaxin-1 with 4. Discussion age and APOε4 optimally differentiated AD and MCI (AUC 0.74), correctly predicting 71% of samples with A plasma biomarker or biomarker set that aids early diag- 75% sensitivity and 66% specificity (Fig. 2B; Table 2). FH nosis, stratification, prediction of disease course, or moni- and eotaxin-1 were higher and FB, sCR1, and MCP-1 toring response to therapy in AD is a major unmet need. were lower in AD compared to MCI samples. The model Numerous studies have sought plasma biomarkers relevant replicated in EMIF-AD MBD (AUC 0.67), correctly predict- to AD, and many putative plasma protein biomarkers have ing 61% of samples with 71% sensitivity and 59% speci- been proposed (reviewed in the study by Baird et al. [28]); ficity. In DCR samples, the model performed poorly (AUC however, none has been robustly replicated. Currently, clini- 0.56). cians rely on neuropsychological testing, a time-consuming tool, to diagnose MCI and AD, with confirmation requiring either expensive neuroimaging (MRI or PET scanning) or 3.2.3. MCI from CTL invasive lumbar puncture to measure CSF markers of amy- The optimal model to differentiate MCI from CTL loid or tau pathology. These methods are not suitable either comprised 15 analytes, each providing weak and indepen- for high-volume screening of presymptomatic individuals, A.R. Morgan et al. / Alzheimer’s & Dementia 15 (2019) 776-787 781 Fig. 1. Ten biomarkers associated with diagnosis in the discovery phase. Boxplots for the 10 biomarkers which demonstrated significant differences in con- centrations between diagnostic groups (Kruskal-Wallis). The P values shown are from the Dunn test with Bonferroni correction for pairwise comparisons; bars indicate significant differences. For graphical convenience and better visualization, high outliers were removed from the boxplots, although all are included in the Kruskal-Wallis analysis. Abbreviation: CRP, C-reactive protein. required to identify early disease, or frequent monitoring plasma in the routine context [29]. Recent technological ad- required in assessing response to interventions. Biomarkers vances have improved assay sensitivity, delivering ultrasen- informative in CSF are currently difficult to measure in sitive assays capable of measuring specific amyloid markers 782 A.R. Morgan et al. / Alzheimer’s & Dementia 15 (2019) 776-787 in plasma [7,8,29–31]. Promising as these developments are, ultrasensitive assays require expensive purpose-built equip- ment beyond routine laboratory capacity and currently too costly for large-scale screening. In this study, we set out to identify plasma analyte sets, measurable using simple multiplex enzyme-linked immuno- sorbent assay, that differentiated AD, MCI, and CTL groups. We took as a starting point the powerful multisource evi- dence that inflammation and complement dysregulation were important components of AD pathogenesis [13–17]. In the discovery phase, we used multiplex and singleplex immunoassays to measure 53 proteins relevant to inflammation and complement dysregulation in a large, well-validated cohort, and identify proteins and/or protein sets associated with AD and/or MCI clinical diagnosis. Ten of the 53 proteins were significantly different between groups of different clinical status; a heterogeneous group of analytes including three complement components (C3, C4, C5), two complement regulators (FH, FI), a soluble form of a complement receptor (sCR1), a classical marker of inflammation (CRP), and three chemokines (eotaxin-1, MCP-1, and MIP-1b). Stepwise selection demonstrated strong interdependence between some analytes, anticipated given that all were selected for relevance to complement and/or inflammation; however, several analytes significantly and independently contributed to distinguishing between clinical groups. To identify the most predictive set, models that tested all combinations of analytes and covariables were generated. The best model for AD versus CTL, including analytes sCR1, FB, FH, eotaxin-1, and MCP-1, with covariables age and APOE status, showed an AUC of 0.79 in the discovery cohort, considered “highly predictive” [32]. The best model for AD versus MCI, including analytes sCR1, eotaxin-1, and MCP-1 with covariables age and APOE status, yielded an AUC of 0.74, considered “moder- ately predictive” [32]. Both models were tested in two independent replication cohorts. In the larger of these, EMIF-AD MBD (comprising 867 samples: 88 AD, 425 MCI, 347 CTL), both models Fig. 2. Receiver operating characteristic (ROC) curves for models distin- replicated, AD versus CTL strongly (AUC 0.81), and AD guishing clinical state or predicting progression. ROC curves were gener- versus MCI moderately (AUC 0.67). In the smaller DCR ated representing models which best differentiated AD from controls (A) cohort (105 AD, 69 MCI, 253 CTL), neither model repli- or AD from MCI (B) in the discovery phase and predicted progression or nonprogression in the EMIF cohort (C). In each case, the area under the cated well (AUC 0.58 for AD vs. CTL; 0.56 for AD vs. curve (AUC) for the selected model was calculated, and compared to that MCI). The reasons for failure to replicate in the DCR cohort for the significant covariables alone, age 1 APOE ε4 in (A) and (B), age are unclear; however, this is a relatively small sample set, alone in (C). (A) Shows that a model including FB, FH, sCR1, MCP-1, 60% of which are CTL samples. The strong replication of and eotaxin-1, along with the covariables age and APOE genotype, differen- both models in the larger multicenter EMIF-AD MBD tiated AD and CTL with a predictive power (AUC) of 0.79 (red line), signif- icantly better than the covariables alone (AUC 0.65; blue line). (B) Shows cohort provokes us to suggest that the analytes identified that a model including sCR1, MCP-1, and eotaxin-1, along with the covari- here, perhaps with other promising biomarkers, might pro- ables age and APOE genotype, differentiated AD and MCI with AUC of vide a basis for a focused, relatively simple and inexpensive 0.74 (red line), significantly better than the covariables alone (AUC 0.63; plasma multiplex test that could aid diagnosis. Further blue line). (C) Shows that a model including FB and FH along with age research in large, well-characterized cohorts to replicate, as covariable differentiated MCI progressors and nonprogressors with AUC of 0.71 (red line). The predictive power was significantly greater validate, and extend these findings is needed to deliver a reli- than that obtained using the covariable alone (AUC 0.66; blue line). Abbre- able screening tool. viations: AD, Alzheimer’s disease; APOE, apolipoprotein E; CTL, control; With the exception of FB, each of the analytes selected MCI, mild cognitive impairment. in the models has previously been associated with AD. A.R. Morgan et al. / Alzheimer’s & Dementia 15 (2019) 776-787 783 Table 2 Multivariate models for distinguishing between diagnostic groups AD vs. CTL AD vs. MCI Predictor LogOR (95% CI) P value LogOR (95% CI) P value Intercept 213.49 (227.16; 0.17) .05 23.62 (27.89; 0.65) .10 Age 0.07 (0.04; 0.12) .00005 0.06 (0.02; 0.10) .002 1 APOE ε4 0.74 (0.22; 1.25) .005 0.41 (20.10; 0.92) .12 2 APOE ε4 2.03 (1.0; 3.05) .0001 1.99 (0.86; 3.13) .0006 Eotaxin-1 1.56 (0.78; 2.35) .00009 1.74 (0.97; 2.52) .00001 MCP-1 21.31 (22.21; 0.40) .0005 21.91 (22.82; 21.01) .00003 sCR1 20.90 (21.85; 0.06) .067 21.36 (22.30; 20.41) .005 FH 2.85 (1.42; 4.27) .00009 n/a n/a FB 22.33 (23.60; 21.06) .0003 n/a n/a Abbreviations: AD, Alzheimer’s disease; APOE, apolipoprotein E; CTL, control; MCI, mild cognitive impairment; logOR (95% CI), log odds ratio of the predictor and their 95% confidence interval; Intercept, log odds ratio if the predictors are equal to 0; 1 APOE E4/2 APOE E4: log odds ratio of possessing 1 or 2 ε4 alleles compared to possessing no ε4 allele; n/a, predictors not included in the given model. NOTE. The table summarizes the selected logistic regression models derived from the AddNeuroMed discovery cohort, AD versus CTL in the left panel, AD versus MCI in the right panel. sCR1 (reduced in AD vs. CTL and MCI) had not been been independently replicated. Of the cohorts available to measured in AD plasma previously but was reported higher us, only EMIF-AD MBD included data on progression of in CSF in AD versus CTL [33]. FH (increased in AD vs. MCI cases to AD; 19% of informative MCI cases had pro- CTL) was reported higher in AD plasma in several studies gressed to AD a year after sampling. Of the 10 analytes [4,34,35], although some reported no difference between measured, two were significantly different; FB levels were clinical groups [36]. Eotaxin-1 (higher in AD plasma vs. higher and FH lower in MCI progressors versus nonprogres- CTL and MCI) and MCP-1 (lower in AD plasma vs. CTL sors. These two biomarkers together with age (the only sig- and MCI), both C–C chemokine family members, were re- nificant covariable) predicted MCI conversion with AUC ported as plasma markers of AD status in several studies 0.71. Notably, FB is a key enzyme in the complement ampli- [37–42]; elevated MCP-1 and eotaxin-1 correlated with fication loop while FH is the critical loop regulator; greater memory impairment in MCI/AD [43]. increased FB and decreased FH seen in progressors would Several studies have reported plasma biomarkers predic- favor amplification, suggesting that amplification loop dys- tive of MCI progression to AD. An 18-analyte biomarker regulation might predispose to progression. We were unable signature dominated by cytokines/chemokines predicted to replicate this finding in other cohorts as data on conver- progression within 5 years with 81% accuracy [44]. A 60- sion were not available. Although the model reported for analyte set was predictive of MCI progression to AD with predicting MCI conversion differs from our previous report 79% accuracy [45], and a 10-analyte panel, including com- [26], both identified markers of complement activation/regu- plement and inflammatory proteins, predicted MCI progres- lation, implying that complement dysregulation is a critical sion to AD with 87% accuracy [22]. Our published study predictor of progression. This finding resonates with preclin- identified a model comprising three analytes, FI, TCC, and ical data suggesting that complement and microglial activa- clusterin that, with APOε4 status, predicted progression tion play important roles as mediators of neurotoxicity in (AUC 0.86) [26]. To date, none of these findings have AD [46]. Further research to replicate and validate markers of complement dysregulation as predictors of progression is required. Table 3 There are limitations to the present study. The cohorts Multivariate model for distinguishing between MCI converters and were collected across a wide range of centers and without nonconverters stringent attention to sampling, separation, and storage pro- Predictor LogOR (95% CI) P value tocols that are important for complement and other immunity Intercept 14.13 (25.77; 34.01) .16 assays; however, despite this suboptimal aspect, character- Age 0.08 (0.04; 0.13) .00019 istic of real-world sample collections, strongly predictive FH 24.15 (26.24; 22.05) .00011 marker sets emerged, increasing the likelihood of utility in FB 2.66 (0.72; 4.60) .0072 clinical practice. For several analytes, the commercial cyto- Abbreviations: MCI, mild cognitive impairment; logOR (95% CI), log kine/chemokine platform was insufficiently sensitive for odds ratio of the predictor and the 95% confidence interval; Intercept, log detection in plasma, highlighting the need for better assays. odds ratio if the predictors are equal to 0. NOTE. The table summarizes the selected logistic regression model For most subjects in the cohorts analyzed, imaging data derived from informative samples in the EMIF cohort for MCI converters and/or CSF samples were not available and thus could not versus nonconverters. be included in the analysis. Despite these limitations, we 784 A.R. Morgan et al. / Alzheimer’s & Dementia 15 (2019) 776-787 discovered and replicated evidence that neuroinflammation Glasgow: Jonathan Cavanagh (PI, EC)3, Catherine Deith*, and complement dysregulation are pathological drivers in Scott Farmer16, John McClean16, Alison McColl3, Andrew AD and thus potential therapeutic targets. Several observa- McPherson*, Paul Scouller*, Murray Sutherland.16 tional studies have reported that long-term use of nonste- Independent advisor: H.W.G.M. (Erik) Boddeke (EC).17 roidal anti-inflammatory drugs is associated with reduced GSK: Jill C. Richardson (EC)18, Shahid Khan11, Phil Mur- risk of dementia [47,48]; however, randomized controlled phy19, Christine A. Parker19, Jai Patel.11 trials and systematic reviews found little or no benefit of Janssen: Declan Jones (EC)6, Peter de Boer4, John Kemp4, nonsteroidal anti-inflammatory drugs [49,50]. Perhaps, Wayne C. Drevets6, Jeffrey S. Nye (deceased), Gayle Wit- interventions in these latter studies were commenced too tenberg6, John Isaac6, Anindya Bhattacharya6, Nick Car- late to confer benefit. Inflammatory biomarkers to stratify ruthers6, Hartmuth Kolb.6 and select patients for targeted early intervention might Kings College London: Carmine M. Pariante (PI)10, Federico benefit future trials of anti-inflammatory interventions. Tar- Turkheimer (PI)20, Gareth J. Barker20, Heidi Byrom10, Diana geting complement dysregulation is, as yet, untested in Cash20, Annamaria Cattaneo10, Antony Gee20, Caitlin Hast- AD. Although current anticomplement drugs are tailored ings10, Nicole Mariani10, Anna McLaughlin10, Valeria Mon- for ultrarare diseases, numerous new drugs are progressing delli10, Maria Nettis10, Naghmeh Nikkheslat10, Karen to the clinic, including for therapy of common inflammatory Randall20, Hannah Sheridan*, Camilla Simmons20, Nisha diseases, for example, age-related macular degeneration Singh20, Victoria Van Loo*, Marta Vicente-Rodriguez20, [51]. Anticomplement drugs designed to access brain and tar- Tobias C. Wood20, Courtney Worrell*, Zuzanna Zajkowska.* geted to preclinical or early MCI patients identified and Lundbeck: Niels Plath (EC)21, Jan Egebjerg21, Hans Eriks- selected using markers of complement dysregulation may son21, Francois Gastambide21, Karen Husted Adams21, offer a new pathway to prevention of AD [52]. Ross Jeggo21, Christian Thomsen21, Jan Torleif Pederson21, Brian Campbell*, Thomas M€oller*, Bob Nelson*, Stevin Zorn.* Acknowledgments University of Texas (subcontracted to Lundbeck): Jason This publication incorporates results from the research proj- O’Connor.22 ect entitled “Wellcome Trust Consortium for Neuroimmu- Oxford: Mary Jane Attenburrow (PI)7,23, Alison Baird, nology of Mood Disorders and Alzheimer’s Disease” Jithen Benjamin23, Stuart Clare25, Philip Cowen7, I-Shu which is funded by a grant from the Wellcome Trust (grant (Dante) Huang24, Samuel Hurley*, Helen Jones23, Simon number: 104025/Z/14/Z). A complete list of Consortium Lovestone7,(AD, PI, EC) Francisca Mada*, Alejo Nevado- members is given in the Annex. Holgado7, Akintayo Oladejo*, Elena Ribe7, Katy Smith23, The AddNeuroMed and DCR plasma samples were provided Anviti Vyas.* by the NIHR Biomedical Research Centre and NIHR De- Pfizer: Zoe Hughes (EC)*, Rita Balice-Gordon*, James mentia Biomedical Research Unit hosted at Kings College Duerr*, Justin R. Piro*, Jonathan Sporn.* London and South London and Maudsley NHS Foundation Southampton: V. Hugh Perry (PI)27, Madeleine Cleal*, Trust and funded by the National Institute for Health Gemma Fryatt27, Diego Gomez-Nicola27, Renzo Man- Research under its Biomedical Research Centers initiative. cuso32, Richard Reynolds.27 Authors’ contributions: B.P.M., S.L., A.R.M., S.T., and Sussex: Neil A. Harrison (PI, EC)28, Mara Cercignani28, Char- A.J.N.-H. contributed to study design. A.R.M., C.L., and lotte L. Clarke28, Elizabeth Hoskins*, Charmaine Kohn*, C.O.’H. processed samples and conducted all assays. S.T. Rosemary Murray*, Lauren Wilcock29, Dominika Wlazly30 and A.J.N.-H. conducted data processing and statistical ana- University of Toronto (sub-contracted to Cambridge): Ho- lyses. B.P.M., A.R.M., S.L., and S.T. contributed to the ward Mount.13 writing of the paper. Other authors contributed to the MD 5 mood disorders workpackages lead; AD 5 EMIF-AD MBD cohort. All authors read and approved the Alzheimer’s disease workpackages lead; PI 5 principal final manuscript. investigator; EC 5 executive committee member. 1 Annex: NIMA—Wellcome Trust Consortium for Neuroim- Department of Psychiatry, School of Clinical Medicine, munology of Mood Disorders and Alzheimer’s Disease. University of Cambridge, CB2 0SZ, UK. 2 Consortium members—Part 1: Cambridgeshire and Peterborough NHS Foundation Trust, Cambridge: Edward T. Bullmore (MD, PI, EC)1,2,11, Junaid Cambridge, CB21 5EF, UK. Bhatti1, Samuel J. Chamberlain1,2, Marta M. Correia1,12, 3 Sackler Centre, Institute of Health & Wellbeing, University Anna L. Crofts1, Amber Dickinson*, Andrew C. Foster*, of Glasgow, Sir Graeme Davies Building, Glasgow, G12 Manfred G. Kitzbichler1, Clare Knight*, Mary-Ellen Ly- 8 TA, UK. nall1, Christina Maurice1, Ciara O’Donnell1, Linda J. Poin- 4 Neuroscience, Janssen Research & Development, Janssen ton1, Peter St George Hyslop1,13,14, Lorinda Turner31, Petra Pharmaceutica NV, Turnhoutseweg 30, B-2340, Beerse, Vertes1, Barry Widmer1, Guy B. Williams.1,14 Belgium. Cardiff: B. Paul Morgan (PI)15, Claire A. Leckey15, Angharad 5 The Maurice Wohl Clinical Neuroscience Institute, Cut- R. Morgan*, Caroline O’Hagan*, Samuel Touchard.15 combe Road, London, SE5 9RT, UK. A.R. Morgan et al. / Alzheimer’s & Dementia 15 (2019) 776-787 785 6 Neuroscience, Janssen Research & Development, LLC, Titusville, NJ, 08560, USA. RESEARCH IN CONTEXT 7 Department of Psychiatry, University of Oxford, Warne- ford Hospital, Oxford, OX3 7JX, UK. 8 1. Systematic review: The authors reviewed the current Brighton & Sussex Medical School, University of Sussex, Brighton, BN1 9RR, UK. literature using traditional (e.g., Google Scholar; 9 PubMed) sources to identify published studies utiliz- Sussex Partnership NHS Foundation Trust, Swandean, BN13 3EP, UK. ing inflammation-relevant plasma biomarkers, in 10 particular complement markers, for diagnosis, stag- Kings College London, Institute of Psychiatry, Psychol- ogy and Neuroscience, Department of Psychological Medi- ing, or risk prediction of Alzheimer’s disease. They cine, London, SE5 9RT, UK. noted the dearth of replicated plasma biomarkers and 11 small sample size in many published studies. Immuno-Psychiatry, Immuno-Inflammation Therapeutic Area Unit, GlaxoSmithKline R&D, Stevenage SG1 2. Interpretation: Our findings identify sets of inflam- 2NY, UK. matory biomarkers in plasma that distinguish clinical 12 MRC Cognition and Brain Sciences Unit, 15 Chaucer subgroups (controls: mild cognitive impairment; Road, Cambridge CB2 7EF, UK. Alzheimer’s disease) in a large multicenter cohort; 13 Tanz Centre for Research in Neurodegenerative Diseases, these replicate in an independent cohort. Markers 60 Leonard Avenue, Toronto, ON M5T 2S8 Canada. predictive of progression were also identified in the 14 Department of Clinical Neurosciences, University of latter cohort. Cambridge, CB2 0SZ, UK. 15 Cardiff University, Cardiff CF10 3AT, UK. 3. Future directions: The findings require further repli- 16 NHS Greater Glasgow and Clyde, 1055 Great Western Rd, cation in additional and larger independent cohorts, Glasgow G12 0XH, UK. before development as a clinically viable multi- 17 University of Groningen, 9712 CP Groningen, Netherlands. plexed test for diagnosis and patient stratification. 18 Neurosciences Virtual PoC DPU, GlaxoSmithKline R&D, Stevenage SG1 2NY, UK. 19 Experimental Medicine Imaging, GlaxoSmithKline R&D, Stevenage SG1 2NY, UK. 20 References King’s College London, Department of Neuroimaging Sciences, Institute of Psychiatry, Psychology & Neurosci- ence, De Crespigny Park, London SE5 8AF, UK. [1] Molinuevo JL, Gispert JD, Dubois B, Heneka MT, Lleo A, 21 Engelborghs S, et al. The AD-CSF-index discriminates Alzheimer’s H. Lundbeck A/S Ottiliavej 9, 2500, Valby, Denmark. 22 disease patients from healthy controls: A validation study. J University of Texas Health Science Center at San Anto- Alzheimers Dis 2013;36:67–77. nio, 7703 Floyd Curl Dr, San Antonio, TX 78229, USA. [2] Bayer AJ. The role of biomarkers and imaging in the clinical diagnosis 23 NIHR Oxford cognitive health Clinical Research Facility, of dementia. Age Ageing 2018;47:641–3. Warneford Hospital, Oxford, OX3 7JX, UK. [3] Ritchie C, Smailagic N, Noel-Storr AH, Ukoumunne O, Ladds EC, 24 Martin S. CSF tau and the CSF tau/ABeta ratio for the diagnosis of The Kennedy Institute of Rheumatology, Roosevelt Dr, Alzheimer’s disease dementia and other dementias in people with Oxford OX3 7FY, UK. mild cognitive impairment (MCI). Cochrane Database Syst Rev 25 Oxford Centre for Functional MRI of the Brain, John Rad- 2017;3:CD010803. cliffe Hospital, Oxford OX3 9DU, UK. [4] Hye A, Lynham S, Thambisetty M, Causevic M, Campbell J, 26 Byers HL, et al. Proteome-based plasma biomarkers for Alzheimer’s Pfizer, Inc, 1 Portland Street, Cambridge MA, USA. 27 disease. Brain 2006;129:3042–50. Centre for Biological Sciences, University of Southamp- [5] Song F, Poljak A, Smythe GA, Sachdev P. Plasma biomarkers for mild ton, Southampton, UK. cognitive impairment and Alzheimer’s disease. Brain Res Rev 2009; 28 Clinical Imaging Sciences Centre (CISC), University of 61:69–80. Sussex, Brighton, BN1 9RR, UK. [6] Bazenet C, Lovestone S. Plasma biomarkers for Alzheimer’s disease: 29 Much needed but tough to find. Biomark Med 2012;6:441–54. Sussex Partnership NHS Foundation Trust, Nevill Avenue, [7] Hanon O, Vidal JS, Lehmann S, Bombois S, Allinquant B, Hove BN3 7HZ, UK. 30 Treluyer JM, et al. Plasma amyloid levels within the Alzheimer’s pro- Brighton & Sussex University Hospitals NHS Trust, cess and correlations with central biomarkers. Alzheimers Dement Brighton BN2 5BE, UK. 2018;14:858–68. 31 Department of Medicine, School of Clinical Medicine, [8] Zetterberg H, Blennow K. From cerebrospinal fluid to blood: The third University of Cambridge, CB2 0SZ, UK. wave of fluid biomarkers for Alzheimer’s disease. J Alzheimers Dis 32 2018;64:S271–9. VIB-KU Leuven Center for Brain & Disease Research, [9] Harold D, Abraham R, Hollingworth P, Sims R, Gerrish A, Campus Gasthuisberg, Herestraat 49, bus 602, 3000 Leuven, Hamshere ML, et al. Genome-wide association study identifies vari- Belgium. ants at CLU and PICALM associated with Alzheimer’s disease. Nat *Former consortium members. Genet 2009;41:1088–93. 786 A.R. Morgan et al. / Alzheimer’s & Dementia 15 (2019) 776-787 [10] Lambert JC, Heath S, Even G, Campion D, Sleegers K, Hiltunen M, [30] Janelidze S, Stomrud E, Palmqvist S, Zetterberg H, van Westen D, et al. Genome-wide association study identifies variants at CLU and Jeromin A, et al. Plasma b-amyloid in Alzheimer’s disease and CR1 associated with Alzheimer’s disease. Nat Genet 2009;41:1094–9. vascular disease. Sci Rep 2016;6:26801. [11] Jones L, Holmans PA, Hamshere ML, Harold D, Moskvina V, [31] Nakamura A, Kaneko N, Villemagne VL, Kato T, Doecke J, Dore V, Ivanov D, et al. Genetic evidence implicates the immune system and et al. High performance plasma amyloid-b biomarkers for Alzheimer’s cholesterol metabolism in the aetiology of Alzheimer’s disease. disease. Nature 2018;554:249–54. PLoS One 2010;5:e13950. [32] Mandic S, Go C, Aggarwal I, Myers J, Froelicher VF. Relationship of [12] Lambert JC, Grenier-Boley B, Chouraki V, Heath S, Zelenika D, predictive modelling to receiver operating characteristics. J Cardio- Fievet N, et al. Implication of the immune system in Alzheimer’s dis- pulm Rehabil Prev 2008;28:415–9. ease: Evidence from genome-wide pathway analysis. J Alzheimers Dis [33] Daborg J, Andreasson U, Pekna M, Lautner R, Hanse E, Minthon L, 2010;20:1107–18. et al. Cerebrospinal fluid levels of complement proteins C3, C4 and [13] International Genomics of Alzheimer’s Disease Consortium IGAP. CR1 in Alzheimer’s disease. J Neural Transm (vienna) 2012; Convergent genetic and expression data implicate immunity in 119:789–97. Alzheimer’s disease. Alzheimers Dement 2015;11:658–71. [34] Honda S, Itoh F, Yoshimoto M, Ohno S, Hinoda Y, Imai K. Association [14] Engelhart MJ, Geerlings MI, Meijer J, Kiliaan A, Ruitenberg A, van between complement regulatory protein factor H and AM34 antigen, Swieten JC, et al. Inflammatory proteins in plasma and the risk of detected in senile plaques. J Gerontol A Biol Sci Med Sci 2000; dementia: the Rotterdam study. Arch Neurol 2004;61:668–72. 55:M265–9. [15] Kuo HK, Yen CJ, Chang CH, Kuo CK, Chen JH, Sorond F. Relation of [35] Gezen-Ak D, Dursun E, Hanagasi H, Bilgiç B, Lohman E, Araz OS, € C-reactive protein to stroke, cognitive disorders, and depression in the et al. BDNF, TNFa, HSP90, CFH, and IL-10 serum levels in patients general population: Systematic review and meta-analysis. Lancet with early or late onset Alzheimer’s disease or mild cognitive impair- Neurol 2005;4:371–80. ment. J Alzheimers Dis 2013;37:185–95. [16] Motta M, Imbesi R, Di Rosa M, Stivala F, Malaguarnera L. Altered [36] Williams MA, Haughton D, Stevenson M, Craig D, Passmore AP, plasma cytokine levels in Alzheimer’s disease: Correlation with the Silvestri G. Plasma complement factor H in Alzheimer’s disease. J disease progression. Immunol Lett 2007;114:46–51. Alzheimers Dis 2015;45:369–72. [17] Aiyaz M, Lupton MK, Proitsi P, Powell JF, Lovestone S. Complement [37] Galimberti D, Schoonenboom N, Scarpini E, Scheltens PDutch-Italian activation as a biomarker for Alzheimer’s disease. Immunobiology Alzheimer Research Group. Chemokines in serum and cerebrospinal 2012;217:204–15. fluid of Alzheimer’s disease patients. Ann Neurol 2003;53:547–8. [18] Kiddle SJ, Sattlecker M, Proitsi P, Simmons A, Westman E, Bazenet C, [38] Galimberti D, Schoonenboom N, Scheltens P, Fenoglio C, et al. Candidate blood proteome markers of Alzheimer’s disease onset Bouwman F, Venturelli E, et al. Intrathecal chemokine synthesis in and progression: A systematic review and replication study. J mild cognitive impairment and Alzheimer disease. Arch Neurol Alzheimers Dis 2014;38:515–31. 2006;63:538–43. [19] Lovestone S, Francis P, Kloszewska I, Mecocci P, Simmons A, [39] Galimberti D, Fenoglio C, Lovati C, Venturelli E, Guidi I, Corra B, Soininen H, et al. AddNeuroMed–the European collaboration for the et al. Serum MCP-1 levels are increased in mild cognitive impairment discovery of novel biomarkers for Alzheimer’s disease. Ann N Y and mild Alzheimer’s disease. Neurobiol Aging 2006;27:1763–8. Acad Sci 2009;1180:36–46. [40] Choi C, Jeong JH, Jang JS, Choi K, Lee J, Kwon J, et al. Multiplex [20] Simmons A, Westman E, Muehlboeck S, Mecocci P, Vellas B, analysis of cytokines in the serum and cerebrospinal fluid of patients Tsolaki M, et al. MRI measures of Alzheimer’s disease and the with Alzheimer’s disease by color-coded bead technology. J Clin Neu- AddNeuroMed study. Ann N Y Acad Sci 2009;1180:47–55. rol 2008;4:84–8. [21] Bos I, Vos S, Vandenberghe R, Scheltens P, Engelborghs S, Frisoni G, [41] Correa JD, Starling D, Teixeira AL, Caramelli P, Silva TA. Chemo- et al. The EMIF-AD Multimodal Biomarker Discovery study: Design, kines in CSF of Alzheimer’s disease patients. Arq Neuropsiquiatr methods and cohort characteristics. Alzheimers Res Ther 2018;10:64. 2011;69:455–9. [22] Hye A, Riddoch-Contreras J, Baird AL, Ashton NJ, Bazenet C, [42] Zhang R, Miller RG, Madison C, Jin X, Honrada R, Harris W, et al. Leung R, et al. Plasma proteins predict conversion to dementia from Systemic immune system alterations in early stages of Alzheimer’s prodromal disease. Alzheimers Dement 2014;10:799–807. disease. J Neuroimmunol 2013;256:38–42. [23] Petersen RC. Mild cognitive impairment as a diagnostic entity. J Intern [43] Bettcher BM, Fitch R, Wynn MJ, Lalli MA, Elofson J, Jastrzab L, et al. Med 2004;256:183–94. MCP-1 and eotaxin-1 selectively and negatively associate with mem- [24] McKhann G, Drachman D, Folstein M, Katzman R, Price D, ory in MCI and Alzheimer’s disease dementia phenotypes. Alzheimers Stadlan EM. Clinical diagnosis of Alzheimer’s disease: report of the Dement 2016;3:91–7. NINCDS-ADRDA work group under the auspices of Department of [44] Ray S, Britschgi M, Herbert C, Takeda-Uchimura Y, Boxer A, Health and Human Services task force on Alzheimer’s disease. Blennow K, et al. Classification and prediction of clinical Alzheimer’s Neurology 1984;34:939–44. diagnosis based on plasma signaling proteins. Nat Med 2007; [25] Morgan AR, Touchard S, O’Hagan C, Sims R, Majounie E, Escott- 13:1359–62. Price V, et al. The correlation between inflammatory biomarkers and [45] Yang H, Lyutvinskiy Y, Herukka SK, Soininen H, Rutishauser D, polygenic risk score in Alzheimer’s disease. J Alzheimers Dis 2017; Zubarev RA. Prognostic polypeptide blood plasma biomarkers of Alz- 56:25–36. heimer’s disease progression. J Alzheimers Dis 2014;40:659–66. [26] Hakobyan S, Harding K, Aiyaz M, Hye A, Dobson R, Baird A, et al. [46] Hansen DV, Hanson JE, Sheng M. Microglia in Alzheimer’s disease. Complement biomarkers as predictors of disease progression in J Cell Biol 2018;217:459–72. Alzheimer’s disease. J Alzheimers Dis 2016;54:707–16. [47] McGeer PL, Rogers J, McGeer EG. Inflammation, anti-inflammatory [27] Morgan AR, O’Hagan C, Touchard S, Lovestone S, Morgan BP. agents, and Alzheimer’s disease: The last 22 years. J Alzheimers Dis Effects of freezer storage time on levels of complement biomarkers. 2016;54:853–7. BMC Res Notes 2017;10:559. [48] Etminan M, Gill S, Samii A. Effect of non-steroidal anti-inflammatory [28] Baird AL, Westwood S, Lovestone S. Blood-based proteomic bio- drugs on risk of Alzheimer’s disease: Systematic review and meta- markers of Alzheimer’s disease pathology. Front Neurol 2015;6:236. analysis of observational studies. BMJ 2003;327:128. [29] Song F, Qian Y, Peng X, Han G, Wang J, Bai Z, et al. Perturbation of [49] Wang J, Tan L, Wang HF, Tan CC, Meng XF, Wang C, et al. Anti-in- the transcriptome: implications of the innate immune system in flammatory drugs and risk of Alzheimer’s disease: An updated system- Alzheimer’s disease. Curr Opin Pharmacol 2016;26:47–53. atic review and meta-analysis. J Alzheimers Dis 2015;44:385–96. A.R. Morgan et al. / Alzheimer’s & Dementia 15 (2019) 776-787 787  [50] Miguel-Alvarez M, Santos-Lozano A, Sanchis-Gomar F, Fiuza-Luces C, [51] Morgan BP, Harris CL. Complement, a target for therapy in inflamma- Pareja-Galeano H, Garatachea N, et al. Non-steroidal anti-inflammatory tory and degenerative diseases. Nat Rev Drug Discov 2015;14:857–77. drugs as a treatment for Alzheimer’s disease: A systematic review and [52] Morgan BP. Complement in the pathogenesis of Alzheimer’s disease. meta-analysis of treatment effect. Drugs Aging 2015;32:139–47. Semin Immunopathol 2018;40:113–24.