Journal of Alzheimer’s Disease 77 (2020) 1353–1368 1353 DOI 10.3233/JAD-200208 IOS Press Dickkopf-1 Overexpression in vitro Nominates Candidate Blood Biomarkers Relating to Alzheimer’s Disease Pathology Liu Shia,∗ , Laura M. Winchestera , Benjamine Y. Liua , Richard Killickb , Elena M. Ribea , Sarah Westwooda , Alison L. Bairda , Noel J. Buckleya , Shengjun Hongc , Valerija Dobricicc , Fabian Kilpertc , Andre Franked , Steven Kiddleb,e , Martina Sattleckerb,e , Richard Dobsonjjj,kkk , Antonio Cuadradog,h , Abdul Hyeb , Nicholas J. Ashtonb,f,i,j , Angharad R. Morgank , Isabelle Bosl,m , Stephanie J.B. Vosl , Mara ten Katem , Philip Scheltensm , Rik Vandenberghen , Silvy Gabeln,o , Karen Meersmansn,o , Sebastiaan Engelborghsp,q,r,t , Ellen E. De Roeckp,q,r , Kristel Sleegersr,s , Giovanni B. Frisoniu,v , Olivier Blinw , Jill C. Richardsonx , Régis Bordety , José L. Molinuevoz,aa , Lorena Ramiaa , Anders Wallini,lll , Petronella Kettuneni,bb , Magda Tsolakicc , Frans Verheyl , Alberto Lleódd , Daniel Alcoleadd , Julius Poppee,ff , Gwendoline Peyratoutee , Pablo Martinez-Lagegg , Mikel Taintagg,hh , Peter Johannsenii , Charlotte E. Teunissenjj , Yvonne Freund-Levikk,ll,mm,nn , Lutz Frölichoo , Cristina Legido-Quigleypp,qq , Frederik Barkhofrr,ss , Kaj Blennowi,tt , Katrine Laura Rasmussenuu,vv,ww , Børge Grønne Nordestgaardvv,ww,xx,hhh , Ruth Frikke-Schmidtuu,vv,ww , Sune Fallgaard Nielsenvv,ww,xx , Hilkka Soininenyy , Bruno Vellaszz , Iwona Kloszewskaaaa , Patrizia Mecoccibbb , Henrik Zetterbergi,tt,ccc,ddd , B. Paul Morgank , Johannes Strefferq,eee , Pieter Jelle Visserl,m,fff , Lars Bertramc,ggg , Alejo J. Nevado-Holgadoa and Simon Lovestonea,iii a Department of Psychiatry, University of Oxford, UK b King’s College London, Institute of Psychiatry, Psychology and Neuroscience, Maurice Wohl Institute Clinical Neuroscience Institute, London, UK c Lübeck Interdisciplinary Platform for Genome Analytics, Institutes of Neurogenetics and Cardiogenetics, University of Lübeck, Lübeck, Germany d Institute of Clinical Molecular Biology, Christian-Albrechts-University of Kiel, Kiel, Germany e MRC Social, Genetic and Developmental Psychiatry Centre, King’s College London, UK f NIHR Biomedical Research Centre for Mental Health and Biomedical Research Unit for Dementia at South London and Maudsley NHS Foundation, London, UK g Centro de Investigación Biomédica en Red sobre Enfermedades Neurodegenerativas (CIBERNED), Instituto de Investigación Sanitaria La Paz (IdiPaz), Instituto de Investigaciones Biomédicas Alberto Sols UAM-CSIC, and Department of Biochemistry, Faculty of Medicine, Autonomous University of Madrid, Madrid, Spain h ”Victor Babes” National Institute of Pathology, Bucharest, Romania i Department of Psychiatry and Neurochemistry, Sahlgrenska Academy at the University of Gothenburg, Mölndal, Sweden j Wallenberg Centre for Molecular and Translational Medicine, University of Gothenburg, Gothenburg, Sweden k Dementia Research Institute Cardiff, Cardiff University, Cardiff, UK l Department of Psychiatry and Neuropsychology, School for Mental Health and Neuroscience, Alzheimer Cen- trum Limburg, Maastricht University, Maastricht, the Netherlands m Alzheimer Center, VU University Medical Center, Amsterdam, the Netherlands ∗ Correspondence to: Liu Shi, Department of Psychiatry, Uni- versity of Oxford, UK. E-mail: liu.shi@psych.ox.ac.uk. ISSN 1387-2877/20/$35.00 © 2020 – IOS Press and the authors. All rights reserved This article is published online with Open Access and distributed under the terms of the Creative Commons Attribution Non-Commercial License (CC BY-NC 4.0). 1354 L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood n University Hospital Leuven, Leuven, Belgium o Laboratory for Cognitive Neurology, Department of Neurosciences, KU Leuven, Belgium p Center for Neurosciences, Vrije Universiteit Brussel (VUB), Brussels, Belgium q Department of Biomedical Sciences, University of Antwerp, Antwerp, Belgium r Institute Born-Bunge, University of Antwerp, Antwerp, Belgium s Neurodegenerative Brain Diseases Group, Center for Molecular Neurology, VIB, Belgium t Department of Neurology, UZ Brussel, Brussels, Belgium u University of Geneva, Geneva, Switzerland v IRCCS Istituto Centro San Giovanni di Dio Fatebenefratelli, Brescia, Italy w AIX Marseille University, INS, Ap-hm, Marseille, France x Neurosciences Therapeutic Area, GlaxoSmithKline R&D, Stevenage, UK y University of Lille, Inserm, CHU Lille, France z Alzheimer’s disease & other cognitive disorders unit, Hospital Clínic, Barcelona, Spain aa BarcelonaBeta Brain Research Center, Universitat Pompeu Fabra, Barcelona, Spain bb Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, UK cc 1st Department of Neurology, AHEPA University Hospital, school of Medicine, Aristotle University of Thes- saloniki, Thessaloniki, Makedonia, Greece dd Department of Neurology, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain ee Department of Psychiatry, University Hospital of Lausanne, Lausanne, Switzerland ff Geriatric Psychiatry, Department of Psychiatry, Geneva University Hospitals, and University of Geneva, Geneva, Switzerland gg CITA-Alzheimer Foundation, San Sebastian, Spain hh Organización Sanitaria Integrada Goierri – Alto Urola, Osakidetza, Spain ii Danish Dementia Research Centre, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark jj Neurochemistry Laboratory, dept of Clinical Chemistry, Amsterdam Neuroscience, Amsterdam University Medical Centers, Vrije Universiteit, Amsterdam, the Netherlands kk School of Medical Sciences, Örebro University, Örebro, Sweden ll Department of Neurobiology, Care Sciences and Society, Division of Clinical Geriatrics, Karolinska Institute, Stockholm, Sweden mm Department of Old Age Psychiatry, Psychology and Neuroscience, King’s College London, UK nn Department of Psychiatry, Örebro Universitetssjukhus, Örebro, Sweden oo Department of Geriatric Psychiatry, Zentralinstitut für Seelische Gesundheit, University of Heidelberg, Mannheim, Germany pp Kings College London, London, UK qq The Systems Medicine Group, Steno Diabetes Center Copenhagen, Gentofte, Denmark rr Department of Radiology and Nuclear Medicine, VU University Medical Center, Amsterdam, The Netherland ss UCL Institutes of Neurology and Healthcare Engineering, London, UK tt Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden uu Department of Clinical Biochemistry, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark vv The Copenhagen General Population Study, Herlev and Gentofte Hospital, Copenhagen University Hospital, Herlev, Denmark ww Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark xx Department of Clinical Biochemistry, Herlev and Gentofte Hospital, Copenhagen University Hospital, Herlev, Denmark yy Neurology / Institute of Clinical Medicine, University of Eastern Finland, Kuopio, Finland zz Toulouse Gerontopole University Hospital, Univeriste Paul Sabatier, INSERM U 558, France aaa Medical University of Lodz, Lodz, Poland bbb Section of Gerontology and Geriatrics, Department of Medicine, University of Perugia, Perugia, Italy ccc UK Dementia Research Institute at UCL, London, United Kingdom ddd Department of Neurodegenerative Disease, UCL Institute of Neurology, London, United Kingdom eee UCB, Braine-l’Alleud, Belgium, formerly Janssen R&D, LLC. Beerse, Belgium at the time of study conduct L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood 1355 fff Department of Neurobiology, Care Sciences and Society, Division of Neurogeriatrics, Karolinska Institutet, Stockholm, Sweden ggg Department of Psychology, University of Oslo, Oslo, Norway hhh The Copenhagen City Heart Study, Frederiksberg Hospital, Copenhagen University Hospital, Frederiksberg, Denmark iii Currently at Janssen-Cilag UK, formerly at Department of Psychiatry, University of Oxford, UK at the time of the study conduct jjj Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, UK kkk Institute of Health Informatics, University College London, London, UK lll Memory Clinic at Department of Neuropsychiatry, Sahlgrenska University Hospital, Mölndal, Sweden Accepted 15 July 2020 Abstract. Background: Previous studies suggest that Dickkopf-1 (DKK1), an inhibitor of Wnt signaling, plays a role in amyloid- induced toxicity and hence Alzheimer’s disease (AD). However, the effect of DKK1 expression on protein expression, and whether such proteins are altered in disease, is unknown. Objective: We aim to test whether DKK1 induced protein signature obtained in vitro were associated with markers of AD pathology as used in the amyloid/tau/neurodegeneration (ATN) framework as well as with clinical outcomes. Methods: We first overexpressed DKK1 in HEK293A cells and quantified 1,128 proteins in cell lysates using aptamer capture arrays (SomaScan) to obtain a protein signature induced by DKK1. We then used the same assay to measure the DKK1-signature proteins in human plasma in two large cohorts, EMIF (n = 785) and ANM (n = 677). Results: We identified a 100-protein signature induced by DKK1 in vitro. Subsets of proteins, along with age and apolipopro- tein E ␧4 genotype distinguished amyloid pathology (A + T–N–, A + T + N–, A + T–N+, and A + T + N+) from no AD pathology (A–T–N–) with an area under the curve of 0.72, 0.81, 0.88, and 0.85, respectively. Furthermore, we found that some signature proteins (e.g., Complement C3 and albumin) were associated with cognitive score and AD diagnosis in both cohorts. Conclusion: Our results add further evidence for a role of DKK regulation of Wnt signaling in AD and suggest that DKK1 induced signature proteins obtained in vitro could reflect the ATN framework as well as predict disease severity and progression in vivo. Keywords: ATN framework, Dickkopf-1, replication, SomaScan, Wnt signaling INTRODUCTION atrophy on magnetic resonance imaging (MRI). Dichotomizing these biomarkers as normal or abnor- Alzheimer’s disease (AD) is characterized by mal results in eight ATN profiles; absence of AD the presence of amyloid-␤ (A␤) containing plaques related pathology (A-T-N-); the Alzheimer’s con- and neurofibrillary tangles composed of modified tinuum including indications of amyloid pathology tau protein together with the progressive loss of (A + T–N–, A + T + N–, A + T–N+, and A + T + N+); synapses and eventually neurons [1]. Recently, the and Suspected Non-Alzheimer Pathology (SNAP), National Institute on Aging-Alzheimer’s Associa- or non-amyloid dementia (A–T–N+, A–T + N–, and tion (NIA-AA) proposed a classification system A–T + N+) [2]. for disease—the ATN framework [2]—based on The ATN framework has considerable face valid- three biomarker types where “A” represents amyloid ity and has rapidly found wide acceptance in the pathology, measured by cortical amyloid positron research field. As clinical trials are increasingly tar- emission tomography (PET) ligand binding or low geting a range of pathologies, the ATN framework cerebrospinal fluid (CSF) A␤42 ; “T” represents tau helps to inform participant inclusion and potentially pathology, measured by elevated CSF phosphory- also trial outcomes [3]. Moreover, the framework has lated tau (P-tau) or cortical tau PET ligand binding; predictive validity with, for example, people with and “N” represents neurodegeneration or neuronal A + T + N+ showing faster decline than other cat- injury, measured by elevated CSF total tau (T- egories [4, 5]. The ATN framework is limited by tau), 18 F-fluoro-deoxyglucose (FDG) PET, or brain biomarkers that are either not yet fully qualified or 1356 L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood are relatively invasive and where access can be dif- MATERIALS AND METHODS ficult. A blood-based version of the ATN framework would be of considerable value and recent progress HEK293A cells with overexpression of DKK1 suggests such an objective is realizable. While considerable progress has been made in In order to establish a DKK1-induced protein understanding the formation, and to some extent signature, a HIS-tagged DKK1 cDNA was syn- the effects of, the three pathological processes that thesized (GENEWIZ, UK) and cloned into the define the ATN classification, much less progress has mammalian expression construct pcDNA3.1+ (Invit- been made in determining the mechanistic relation- rogen, UK) and validated by sequencing. HEK293A ship between amyloid and tau pathologies and their cells (an adherent strain of HEK293) were cultured in effects in inducing neuronal dysfunction and death. DMEM + 10% FCS in 12-well plates until 80% con- One potential mechanism that has been proposed to fluent and transfected with the DKK1 construct or the link all three processes is Wnt signaling. Activation empty vector control using FuGene 6 according to the of Wnt signaling is neuroprotective against the toxic- manufacturer instructions (Promega, UK). The next ity of A␤ both in vitro and in vivo [6, 7] and reverses day, the FCS content of the media was adjusted to 2% cognitive deficits in rodent models [8, 9]. Inhibition and the cells maintained for a further 6 h. Media was of Wnt signaling could therefore be a factor triggering then removed and total cell lysates collected in M- the onset and progression of AD (reviewed in [10]). PER (ThermoFisher, UK) for proteomic array studies In line with this, Dickkopf-1 (DKK1), an inhibitor using the SOMAScan platform (SomaLogic, Boul- of Wnt signaling, has been reported to be elevated der, CO), allowing for the simultaneous measurement in human AD brain, as well as in mouse models and quantification of 1,128 proteins (n = 5 per condi- with A␤ generation and plaque formation [11–13]. tion). All protein data were log-transformed prior to In model systems, DKK1 is induced by A␤, which analysis. in turn drives synaptic loss, tau phosphorylation, and neuronal death [14–16] and blockade of DKK1 Study participants protects synapses from A␤-mediated neurotoxicity [16, 17]. We used plasma samples recruited from two Collectively, DKK1 mediated inhibition of Wnt previously reported cross European studies: AddNeu- signaling might be a critical factor in the A␤- roMed (ANM) [19] and the European Medical mediated pathway driving tau pathology and hence Information Framework for Alzheimer’s disease neuronal dysfunction and loss. Previously we demon- multi-modal biomarker discovery (EMIF-AD MBD) strated that knockdown of genes on a shared study [20]. A␤/DKK1 pathway also protected neurons from A␤- ANM sample collection was performed at six induced toxicity and that in mice, overexpression of different centers across Europe: University of Kuo- DKK1 induced tau pathology and cognitive impair- pio, Finland; Aristotle University of Thessaloniki, ment [18]. Given this, we propose that the DKK1 Greece; King’s College London, United Kingdom; induced pathway might reflect the ATN framework in University of Lodz, Poland; University of Perugia, man. However, the molecular signature we previously Italy; and University of Toulouse, France [19]. We identified was based on neuronal gene expression used 677 subjects from the ANM cohort includ- and included many master regulators or transcrip- ing 319 AD patients, 149 MCI individuals, and tion factors and hence was not readily translatable 209 elderly unaffected controls (CTL). General clin- to human studies. Therefore, in order to explore ical and demographic information were available whether a DKK1 induced signature was recogniz- for all subjects (including APOE ␧4 genotype data) able in peripheral fluids in human in vivo studies, we (Supplementary Table 1). Furthermore, the majority first determined a protein signature induced by DKK1 participants (84%) had Mini-Mental State Examina- in human cells in vitro and then tested whether this tion (MMSE) measurement and around 60% of the empirically defined protein signature was associated AD patients had Alzheimer’s Disease Assessment with the ATN framework and clinical outcomes in Scale - Cognitive subscale (ADAS-Cog) measure- human plasma from two large independent cohorts ment (Supplementary Table 1) [21]. including people clinically defined with AD, with The EMIF-AD MBD is part of the European mild cognitive impairment (MCI) and apparently Medical Information Framework for Alzheimer’s healthy controls (Fig. 1). disease (EMIF-AD; http://www.emif.eu/), a Euro- L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood 1357 Fig. 1. Flowchart of study design. 1) Measurement and quantification of 1,128 proteins in total cell lysates of HEK293A cells overexpressing DKK1; 2) The top 100 proteins that constitute the DKK1-induced signature were identified using partial least squares (PLS) regression; 3) Measuring the association of 100 DKK1-induced proteins with ATN framework in vivo and obtaining 23 proteins that were significantly associated with any single ATN abnormal; 4) Measuring the association of AD related proteins with other AD clinical outcomes; 5) Mendelian randomization to explore the causal relationship between two proteins (complement component 3 [C3] and fibrinogen gamma chain [FGG]) and AD risk, amyloid and tau (both T-tau and P-tau) pathology. AD, Alzheimer’s disease; SNAP, Suspected Non-Alzheimer Pathology; T-tau, total tau; P-tau, phosphorylated tau. Table 1 Characteristics of 785 participants split by ATN framework. Standard deviation is shown in brackets for age and MMSE in each category. Percentage of cases is shown in brackets for male sex as well as APOE ␧4 carriers. p-values compare each demographic across 8 categories. NPC, No Pathology Control; SNAP, Suspected Non-Alzheimer Pathology; SD, standard deviation; MMSE, Mini-Mental State Examination Variable NPC AD SNAP p A–T–N– A + T–N– A + T + N– A + T–N+ A + T + N+ A–T–N+ A–T + N– A–T + N+ N (total = 785) 229 105 19 54 298 26 18 36 NA Age (y) (SD) 64 (8.5) 68 (9.7) 72 (6.8) 72 (7.3) 70 (8.0) 72 (7.1) 64 (6.3) 71 (8.2) <0.001 Male sex N (%) 123 (54) 57 (54) 9 (47) 30 (56) 170 (57) 14 (54) 10 (56) 23 (64) 0.95 APOE ␧4 carriers N (%) 52 (23) 58 (55) 13 (68) 36 (67) 193 (65) 5 (19) 5 (28) 9 (25) <0.001 MMSE (SD) 27.7 (2.5) 26.5 (3.9) 25.4 (3.9) 24.4 (4.3) 24.0 (4.4) 26.7 (3.6) 27.8 (1.7) 26.9 (2.8) <0.001 pean wide collaboration to facilitate the re-use of demographic information, each subject had a mea- existing healthcare data and the sharing of cohort sure of both A␤ and tau (including total tau [T-tau] samples for the benefit of AD research. Overall, the and phosphorylated tau [P-tau]) pathology. The clas- EMIF-AD MBD study collected samples from 11 sification of the status (abnormal/normal) of amyloid, European cohorts DESCRIPA, EDAR, PharmaCog, T-tau, and P-tau has been described previously [20]. Amsterdam, Antwerp, San Sebastian GAP, Gothen- Briefly, amyloid pathology was determined using burg, Barcelona IDIBAPS, Lausanne, Leuven, and CSF A␤42 in the majority and PET amyloid in a Barcelona St Pau [20]. We used 785 subjects from minority, based on which the individuals were classi- the EMIF study comprising 183 AD patients, 382 fied into abnormal and normal status [20]. The levels MCI, and 220 CTL. In addition to general clinical and of T-tau and P-tau in CSF were analyzed locally and 1358 L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood the local cut-off point was used to determine their The top 100 proteins that contributed to this multi- status (abnormal/normal) [20]. For these subjects, variate signature most constitute the ‘DKK1-induced “A” is defined by amyloid status, “T” is by P-tau signature’ for subsequent analysis. The biological status, and “N” is by T-tau status. Dichotomizing significance of the DKK1-induced signature was these biomarkers as normal or abnormal results in evaluated using the DAVID Bioinformatics Resource, eight ATN profiles (Table 1). In addition, each sub- version 6.8 Functional Annotation tool. Briefly, the ject had MMSE measurement and the majority (over 100 proteins were selected as our ‘gene list’ while all 72%) had other neuropsychological tests for memory, 1,128 proteins quantified in the study were selected as language, and executive functioning as previously our ‘background gene list’. The enrichment analysis reported (Supplementary Table 1) [20]. Furthermore, was performed on the KEGG database. each individual had genome-wide single nucleotide To compare the association of proteins with the polymorphism (SNP) genotyping. The details of SNP ATN framework, we used logistic regression to assays and raw data processing were described in compare proteins in different ATN profiles to ‘no [22]. pathology controls’ (A–T–N–), adjusting for age, APOE ␧4 genotype, and gender. Logistic regres- sion was also used to compare proteins in different Protein quantification of human plasma in two AD diagnostic groups as well as between MCI par- cohorts ticipants who subsequently converted to dementia (MCIc) within 3 years relative to those whose MCI The SOMAScan assay, an aptamer-based assay remained stable (MCIs). To analyze the association (SomaLogic, Boulder, CO) [23], was used to mea- of proteins with continuous AD phenotypes (i.e., sure plasma proteins in subjects collected from both MMSE), we used partial correlation and adjusted ANM and EMIF cohorts. Because of an evolving plat- for age, APOE ␧4 genotype, and gender. p values form, different versions of the SOMAScan assay were obtained from both logistic regression and partial used in samples from the two cohorts, with 1,016 correlation were corrected using false discovery rate and 4,001 proteins measured in ANM and EMIF (FDR) and corrected p values were presented in cohorts, respectively. The in vitro experiments were heat map. Furthermore, for visual presentation, we conducted with a version of the assay having 1,128 presented the absolute protein expression value in dif- proteins. The three versions of the assay used here ferent ATN and AD diagnostic groups in box plots. were generated such as to ensure data interoperability. Student’s t-test was used to assess pairwise difference The abundance of each protein was log-transformed and uncorrected p values were presented in the box for all subsequent analyses. plots. Forward stepwise logistic regression was used to Statistical analysis find the analyte set that optimally discriminated amy- loid pathology (A + T–N–, A + T + N–, A + T–N+, All statistical analyses were completed using R and A + T + N+) from no AD pathology (A–T–N–) in (version 3.3.2). We used Student’s t-test to assess all subjects as well as in only cognitively normal indi- the relative levels of DKK family proteins (DKK1, viduals. In both cases, demographic covariates age DKK3, DKK4, and DKK-Like 1 [DKKL1]) as and APOE ␧4 genotype were included in models as measured on the SomaLogic platform, in DKK1 over- potential predictors. For each comparison, the data set expressing and control cells. Partial least squares was randomly split into training (90%) and validation (PLS) regression was used to obtain a signature con- (10%) sets. The training set was used to select vari- sisting of the 100 top, or most altered, proteins as a ables and fit the model which was then tested on the consequence of DKK1 expression in HEK293A cell validation set using receiver operating curve (ROC) lysates. A PLS regression model was fitted to the data analysis. The 95% confidence Intervals of AUC was using all 1,128 proteins as the predictors (X) and the calculated using the ci.auc function. DKK1 or control status as the response variable (Y). We ranked proteins based on the calculated coeffi- Mendelian randomization cients using two components from the resulting PLS regression model. The coefficients corresponding to Mendelian randomization (MR) was used to inves- each protein in the model are a proxy for how much tigate the causal relationship between two most each protein contributes to the signal. promising (see Results section) proteins (C3 and L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood 1359 FGG) and AD risk, A␤ and tau (T-tau and P- markedly from zero, it indicates that substantial hor- tau) pathology. As genetic variants are passed from izontal pleiotropy of the SNPs is less likely [32]. parents to child at conception and remain largely Furthermore, to determine if there was any sin- unaltered by environment throughout an individual’s gle SNP driving the relationship, we performed a lifetime, reverse causation and confounding can be leave-one-out analysis where the MR is performed limited, making MR a powerful tool to examine removing a different SNP in each iteration [33]. causality between the exposure and outcome [24, 25]. The MR approach was based on three assumptions: RESULTS 1) the genetic variants used as instrumental variables are associated with exposures; 2) the genetic vari- DKK1 and DKK4 were differentially expressed in ants are not associated with any confounders of the DKK1-overexpressing cells compared to controls exposure-outcome relation; 3) the genetic variants are associated with outcome only through the exposure, In addition to a large number of other proteins rep- namely a lack of pleiotropy [24]. resenting a wide range of biological processes, both For C3, we selected three SNPs as instrumen- versions of the SOMAScan assay used here include tal variables; rs1065489 [CFH], rs429608 [SKIV2L], aptamers selected to bind and hence measure rela- and rs448260 [C3]. The association of these SNPs tive abundance of DKK proteins including DKK1, with plasma C3 levels have been validated in 95,442 DKK3, and DKK4 and the related protein DKK- individuals (p < 10–67 ) [26]. For FGG, we selected Like 1 (DKKL1). In order to determine the protein 24 SNPs as instrumental variables. These have been signature induced by DKK1, we used the SOMAS- shown to be significantly associated with plasma can assay to compare lysates from HEK293A cells FGG levels in a large genome wide association engineered to over-express human DKK1 with con- studies (GWAS) study including more than 100,000 trol cells transfected with empty vector (n = 5 in subjects [27] (assumption 1). Then we checked each case). We first determined the ability of the whether the SNPs were either in linkage disequi- SOMAScan assay to detect DKK1 and to differen- librium with one another, or were associated with tiate this protein from other structurally similar DKK known risk factors for AD (e.g., APOE ␧4 genotype) isoforms. We found that among the DKK family pro- (assumption 2). After verifying no direct association teins, both DKK1 and DKK4 registered an increase with AD, we acquired the summary statistics of each in DKK1-overexpressing cells compared both to the SNP with C3 and FGG from both studies separately as other DKK isoforms and to all other protein measures exposure estimates [26, 27]. For AD risk as outcome, (for DKK1 p = 0.008, corrected p = 0.7; for DKK4 we acquired the summary statistics for the associa- p = 0.008, corrected p = 0.7) (Fig. 2). The most likely tion of each SNP with AD risk from a previously explanation for this observation is that the DKK4 published GWAS study; International Genomics of read-out on the SomaLogic panel is in fact at least Alzheimer’s Project (IGAP) by Lambert et al. [28]. in part a read-out of DKK1 due to cross reactivity One C3 SNP (rs429608 [SKIV2L]) was not found and hence we refer subsequently to this as DKK1/4. in IGAP; we therefore acquired summary statistics of rs429608 in another GWAS study - UK Biobank DKK1-induced proteomic signature was (UKBB) [29]. For A␤ and tau (T-tau and P-tau) as enriched in AD pathways outcome, we analyzed the association of each SNP with A␤ and tau status in EMIF cohort using PLINK Having shown that the SOMAScan assay identi- (v1.7). Using a two-sample MR approach, the expo- fies the overexpression of DKK1 (with concomitant sure SNP (SNP-C3 and SNP-FGG) and outcome SNP signal in the aptamer raised against DKK4), we (SNP - AD risk, SNP - A␤, SNP - T-tau, and SNP - used PLS to identify a multivariate proteomic sig- P-tau) associations were used to compute estimates nature that distinguished DKK1 overexpression cell of each exposure-outcome association. We then used lysates from controls. We ranked the proteins based two MR methods to test the robust causal infer- on PLS coefficients (absolute value), and then ence including an inverse-variance weighted (IVW) selected the 100 proteins with the largest contri- method [30] the weighted median method [31]. bution to the DKK1-induced multivariate signature To test the third assumption, we used the MR- (Supplementary Table 2). As an exploratory study Egger method to calculate values of intercepts and only we then assessed the biological significance their p values. If the intercepts do not deviate of this signature using the DAVID Bioinformatics 1360 L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood Fig. 2. DKK1 overexpression leads to higher levels of (A) DKK1 and (B) DKK4 expression in HEK293A cell lysate (n = 5 per condition). Y axis represents the log transformed of proteins expression abundance measured by Somascan assay. Resource. Overall, eight pathways were enriched the SNAP group (A– but T+ or N+). Based on though not passing multiple correction (Supplemen- their expression, these proteins could be divided into tary Table 3). Interestingly, AD was nominated as the three groups: 1) those influenced only by amyloid second highest-ranked pathway (p = 0.014, corrected pathology (A+) and independent of P-tau or T-tau p = 0.075) in relation to disease. status. For example, FGG increased in all individuals with abnormal amyloid pathology although it did not Association of DKK1-induced signature with the achieve significance in A + T + N– subjects (Fig. 3B), ATN framework in vivo perhaps due to a low number of only 19 subjects with this profile. Other proteins belonging to this group We then investigated whether the top 100 pro- were BRF-1, Coagulation Factor VII, CKAP2, HMG- teins induced by DKK1, referred to here as the 1, CAMK2D, AURKB, BFL1, C3, and albumin; 2) DKK1-induced signature, was associated with the those influenced by both amyloid and T-tau (A + N+). ATN framework in the EMIF cohort (n = 785). Table 1 For example, DKK1/4 increased in individuals with shows the characteristics of the participants split by A + T–N+ and A + T + N+ profiles (Fig. 3C). Three ATN framework. No significant difference in sex was other proteins also belonged to this group: eotaxin, found among 8 profiles. Participants within the AD coactosin-like protein, and annexin I; 3) those influ- and SNAP groups were older than those in the ‘No enced by amyloid, P-tau, and T-tau (A + T + N+), Pathology Control’ (NPC) group except those with an resulting changed levels in only A + T + N+ individu- A–T + N– profile. The prevalence of APOE ␧4 carri- als, e.g., DKK1 (Fig. 3D), CHST2, MK01, CONA1, ers was higher in AD group than those in NPC or DLRB1, FN1.4, Cytochrome c, and SHP-2. SNAP. Furthermore, MMSE was lower in AD group We then used forward stepwise logistic regres- than individuals in NPC or SNAP except those in sion to identify optimal analyte sets to distinguish A + T–N– profile. the NPC group from different ATN profiles within We used logistic regression to compare proteins the AD (A+) group. We used AD related proteins in different ATN framework groups to the NPC as well as age and APOE ␧4 genotype as input group (A–T–N–), adjusting for age, APOE ␧4 geno- features. Results showed that a model containing type, and gender. Of the 100 proteins, the levels of HMG-1 as well as age and APOE ␧4 genotype 23 proteins significantly altered in participants with best discriminate A + T-N- from NPC group with at least one abnormal ATN biomarker, i.e., either an area under the curve (AUC) of 0.72 (Fig. 3E). A+, T+ or N+ after FDR correction (Fig. 3A). Fur- The optimal model to differentiate A + T + N– from thermore, the majority of proteins were associated NPC group contained 6 features including SHP-2, with amyloid-related pathology rather than with non- FN1.4, CKAP2, and CHST2 as well as age and amyloid pathology (SNAP). In subsequent analyses APOE ␧4 genotype. For differentiating A + T–N+ we therefore focused on comparisons within the AD from NPC group, a model containing 7 proteins (A+) group to the NPC (A–T–N–) group, omitting (annexin I, Albumin, Cytochrome c, Eotaxin, DKK1, L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood 1361 Fig. 3. A) Association of 23 DKK1-induced signature with 7 ATN profiles compared to A–T–N–. B–D) Comparison of proteins between A–T–N– (n = 229) and amyloid-positive individuals including A + T–N– (n = 105), A + T + N– (n = 19), A + T–N+ (n = 54), and A + T + N+ (n = 289). E, F) AUC of using proteins along with age and APOE ␧4 genotype to differentiate A–T–N– from amyloid-positive individuals in all individuals and healthy controls respectively. High and low beta indicate positive and negative coefficients respectively. SNAP, Suspected Non-Alzheimer Pathology; FGG, fibrinogen gamma chain. In B, C, and D, Y axis represents the log transformed of proteins expression abundance measured by Somascan assay. ∗ p < 0.05; ∗∗∗ p < 0.001; NS., not significant; AUC, area under the curve; CTL, controls. 1362 L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood Fig. 4. A) Comparison of C3 in different AD diagnostic groups in both EMIF and ANM cohorts. B) comparison of C3 in MCI who subsequently converted to dementia (MCIc) to those whose MCI remained stable (MCIs) in both cohorts. C) Correlation of MMSE with C3 in both cohorts. Y axis in A & B and X axis of C represent the log transformed of proteins expression abundance measured by Somascan assay. C3, complement component 3; ANM, AddNeuroMed; EMIF, European Medical Information Framework. CONA1, and AURKB) with age and APOE ␧4 geno- in the EMIF cohort, few of these replicated in the type was selected. The optimal model to differentiate ANM cohort (Supplementary Figure 1). However, A + T + N+ from NPC group included 7 features we did find robust replication for two protein asso- including C3, AURKB, Eotaxin, BRF-1, and albu- ciations with clinical features; C3 and albumin. In min as well as age and APOE ␧4 genotype (Fig. 3E). the EMIF cohort, C3 was significantly decreased in The comparison between each optimal model and AD compared to CTL and MCI individuals, a change the combination of age and APOE ␧4 genotype were also observed in the ANM cohort (Fig. 4A). Fur- shown in Supplementary Table 4. thermore, C3 was significantly decreased in MCIc We also set out to determine the optimal ana- (EMIF n = 100; ANM n = 43) relative to MCIs (EMIF lyte sets to differentiate the NPC group from the n = 219; ANM n = 106) in both two cohorts (Fig. 4B). amyloid positive ATN profiles (A+) groups in only C3 had consistent protective effects on cognition as cognitively normal individuals (Fig. 3F). The opti- it was positively associated with MMSE score in mal model that separated the A + T + N+ from the both EMIF (R2 = 0.29, p < 0.001) and ANM cohorts NPC group included four features including CKAP2, (R2 = 0.08, p < 0.001) although the association was C3, age, and APOE ␧4 genotype. A model consist- weak in the ANM cohort (Fig. 4C). Albumin was ing of four proteins (Coagulation Factor VII, SHP-2, also significantly decreased in AD compared to CTL FN1.4, and DKK4) best discriminated the A + T–N– in both cohorts (Supplementary Figure 2A). Fur- group from the NPC group with an AUC of 0.92 and thermore, it was positively associated with baseline a model consisting of three proteins (CKAP2, FN1.4, MMSE score in both EMIF (R2 = 0.25, p < 0.001) and and Cytochrome c) together with age and APOE dif- ANM cohorts (R2 = 0.17, p < 0.001) (Supplementary ferentiated the A + T–N+ group from the NPC with an Figure 2B). AUC of 0.94 (Fig. 3F) and (Supplementary Table 4). The A + T + N– group, with only three members, was Causal relationship between C3 and AD risk, too small to test for differentiation from the NPC amyloid and tau pathology group. We then sought to investigate the causal relation- Association between ATN related proteins and ship between two proteins (C3 and FGG) with AD AD clinical outcomes risk and A␤ and tau pathology (T-tau and P-tau) using two sample MR. A prerequisite for MR is evidence of We then determined the relationship between ATN genetic variations associated with the exposure vari- framework-related proteins and AD clinical out- able, in this case levels of the proteins in plasma. To comes in two large independent cohorts: EMIF (183 identify such variants, we interrogated the GWAS cat- AD, 382 MCI, and 220 CTL) and ANM (319 AD, alogue for all genes encoding the proteins in the panel 149 MCI and 209 CTL). While we found many pro- and found that four proteins with at least two SNPs teins to be associated with AD clinical outcomes in their encoding gene significantly associated with L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood 1363 their levels in blood (p < 10–8 ). These proteins were identified a DKK1-induced signature from in vitro C3, FGG, CONA1, and coagulation factor VII. Given human cell models of DKK1 overexpression and then that C3 and FGG were associated with AD from pre- determined the association of this signature with AD vious biomarker studies from our group and others pathology in two large independent cohorts; EMIF [26, 34–36], we selected these two proteins to further (n = 785) and ANM (n = 677). From analysis of high explore their causal relationship with AD. For expo- dimensionality proteomics of over 1,000 proteins, sure estimates, we selected three C3 SNPs [26] and we determined a 100-protein signature induced by 24 FGG SNPs [27] as instruments for MR analysis. DKK1 and found that this protein set was enriched in For AD risk as outcome, we acquired the summary molecular pathways known to be associated with AD, statistics for the association of each SNP with AD risk adding further evidence to the relevance and possible from IGAP and UK Biobank [28, 29]. For A␤ and tau importance of this mechanism or pathway in disease. pathologies as outcomes, we obtained the association We then explored the relationship of the identified of each SNP with biomarker-based A␤ and tau status signature to the biomarker based AD classification, from the EMIF cohort. i.e., ATN framework. The role of DKK1 and Wnt We first confirmed no pleiotropic effects for signaling in AD is suggested, from multiple lines these genes given that intercept of genetic variants of evidence mentioned above, to somehow trans- from MR-Egger regression was close to zero. We mit a signal from amyloid to tau pathology and then performed MR analysis using both weighted hence neurotoxicity [14–16]. The mechanism of median (WM) and inverse-variance weighted (IVW) such a transmission is unknown although might be approaches. Results showed that lower C3 was likely through the canonical Wnt pathway regulation of the to be causally related to high AD risk using both WM tau kinase glycogen synthase kinase-3 (GSK3) or (effect size [␤] = 0.75, standard error of the effect size through the non-canonical Wnt pathways that include [se] = 0.44, 95% CI [–0.14, 1.63], p = 0.09) and IVW Rho/Roc and their effects on synaptic resilience [10, (␤ = 0.72, se = 0.40, 95% CI [–0.05, 1.50], p = 0.06) 40, 41]. If DKK1 is either a direct or indirect link methods (Fig. 5). In contrast, such a relationship was between amyloid and tau pathology with subsequent not found between C3 with either A␤ or tau (both effects on neurotoxicity, one would expect that a T-tau and P-tau) status (Fig. 5). Furthermore, results DKK1 induced signature might be associated with from the leave-one-out analysis demonstrated that no these different components of the AD pathway. We single SNP was driving the majority of the associa- indeed found this to be the case with a strong associ- tion signal between C3 and AD risk (Supplementary ation of many of the DKK1-signature proteins being Figure 3). For FGG, no causal relationship was found associated with ATN classifications including amy- between FGG and AD risk, or FGG and A␤ or FGG loid (i.e., AD group, A+) but less association with the and tau (Supplementary Figure 4 and Supplementary non-amyloid group (SNAP, A–). These results offer Table 5). confirmation of DKK inhibition of Wnt signaling as a factor in Alzheimer’s pathology and specifically add weight to data from in vitro and in vivo models and DISCUSSION from human brain studies that DKK1 is increased in response to amyloid and as a consequence increases Identification of biomarkers tends to fall into two risk of tau pathology and neurodegeneration. different designs: either hypothesis-driven targeted Based on this, we further investigated the pre- measures of features known to be associated with the dictive value of DKK1-induced proteins in discrim- disease in question or data-driven high dimensional- inating amyloid pathology (A + T–N–, A + T + N–, ity agnostic platform approaches (“omics”). Here we A + T–N+, and A + T + N+) from no AD pathology in utilize a novel approach with a hybrid design where all subjects as well as in cognitively normal individ- first we used agnostic high dimensionality proteomics uals. We found the combination of different subsets in an in vitro model of a hypothesized driver of disease of proteins along with age and APOE ␧4 genotype mechanism and then used the derived signature in a was able to differentiate the different ATN profiles targeted study in human samples. As Wnt signaling with a high AUC, especially in normal individuals has been proposed to be protective and an increase (AUC > 0.9). It should be noted that forward stepwise in the Wnt inhibitor DKK1 has been found to be regression makes an arbitrary decision as to select increased in AD [10, 37–39], and hence a possible highly correlated proteins. Therefore, other proteins driver of disease mechanisms, we first empirically that were highly correlated to those selected proteins 1364 L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood Fig. 5. Forest plot of Mendelian randomization estimates the effects of C3 on AD risk, A␤ and tau status (T-tau and P-tau). Lower C3 is likely to be causally related to high AD risk from both inverse-variance weighted (IVW) and weighted median (WM) methods, but such relationship was not found between C3 and amyloid or C3 and tau (T-tau and P-tau). ∗ rs448260 was not found in EMIF data, therefore its proxy rs2287848 (r2 = 0.93) was used to obtain its association with amyloid and tau. EMIF, European Medical Information Framework; T-tau, total tau; P-tau, phosphorylated tau. could equally function as biomarkers for ATN clas- and in sample size. As the data was derived from sification. For example, HMG-1 was selected in the a biomarker platform with claims to have value model to discriminate A + T–N– from NPC in all sub- in other clinical settings (see for example Soma- jects. As shown in Supplementary Figure 5, it was logic.com/somasignals), then the identification of a highly related to BRF.1, indicating that BRF.1 could signature indicative of AD pathological processes also be a useful marker to discriminate A + T–N– might have value in screening from existing data from NPC. for possible suitability for clinical trials or, when Our study is the largest we are aware of to report therapies become available, possibly for early inter- a plasma biomarker indicative of the ATN frame- vention, including those related to this particular work both in terms of the number of proteins assayed target. As an approach to precision medicine, this L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood 1365 model of biomarker discovery might have wider well-characterized cohorts to replicate, validate, and applicability. extend these findings is needed. In addition to the ATN framework, we also found In conclusion, our results add to the evidence the DKK1-derived protein signature associated with base indicating a role for DKK1 and the Wnt AD clinical outcomes (i.e., MMSE score and MCI pathway in AD pathogenesis. It suggests that a pro- conversion). We observed a robust replication espe- tein signature derived from a human cell model of cially for the association of C3 with clinical features, DKK1 overexpression, when measured in human in line with some recent genetic and biomarkers plasma, is significantly associated with the stag- studies [34, 36, 42]. For example, genetic studies ing according to the ATN framework and could strongly implicate complement signaling with AD discriminate different amyloid-positive classes from pathogenesis with increasing attention being paid No Pathology Controls. Furthermore, a subset of especially to the complement pathway node of C3 the DKK1-signature proteins are strongly associ- and C5 metabolism, regulated by CR1, one of the ated with disease severity and progression and these common variants most strongly associated with AD association can be replicated in two large indepen- [42]. Biomarker studies too suggest that complement dent cohorts. Taken together, our results indicate that signaling is critically altered in AD with a large range this novel, empirically generated approach can help of complement proteins, including C3 being repeat- identify biomarkers of utility for the selection of par- edly nominated in agnostic proteomic studies [34, ticipants for clinical trials as well as for monitoring 36]. However, neither genetic nor biomarker studies trial outcomes. can alone demonstrate directional causality—in other words, although the genetics strongly implicate com- plement as a causative biological process neither they ACKNOWLEDGMENTS nor the biomarker studies are able to say whether any given complement protein is exacerbating or protect- This research was conducted as part of the ing against disease. EMIF-AD project which has received support from Empirical studies have attempted to address this the Innovative Medicines Initiative Joint Under- question, critically important in drug development, taking under EMIF grant agreement n◦ 115372, using model systems. However, the results of such resources of which are composed of financial studies are less than clear. In some preclinical in vivo contribution from the European Union’s Seventh studies, C3 knock out offers synaptic protection and Framework Programme (FP7/2007-2013) and EFPIA reduces amyloid burden in a range of models [43, companies’ in-kind contribution. The DESCRIPA 44], although in other studies increased amyloid accu- study was funded by the European Commission mulation and neurodegeneration has been reported within the 5th framework program (QLRT-2001- [45, 46]. Therefore, to further explore causality, we 2455). The EDAR study was funded by the employed a Mendelian randomization approach with European Commission within the 5th framework this protein data combined with knowledge of C3 program (contract # 37670). The Leuven cohort quantitative trait locus (QTL) SNPs associated with was funded by the Stichting voor Alzheimer Onder- protein levels in this data set. We were able to sub- zoek (grant numbers #11020, #13007 and #15005). stantiate the association between low levels of C3 The Lausanne cohort study was supported by a and high risk of AD, suggesting a causal influence of grant from the Swiss National Research Founda- C3 on risk of AD, further supporting findings from tion to JP (SNF 320030 141179). RV is a senior Rasmussen et al. [26] who studied more than 95,000 clinical investigator of the Flemish Research Foun- individuals from the general population. dation (FWO). JS is currently an employee of We acknowledge that the sample distribution UCB, Braine-l’Alleud, Belgium. The San Sebastian within the ATN framework is a limitation of this GAP study is partially funded by the Department study. As shown in Table 1, A + T + N– profile of Health of the Basque Government (alloca- only included 19 subjects. The small sample size tion 17.0.1.08.12.0000.2.454.01.41142.001.H). We might explain why proteins did not reach signif- acknowledge the contribution of the personnel of icance in A + T + N– profile when comparing to the Genomic Service Facility at the VIB-U Antwerp A–T–N– profile. It might also lead to the fact that Center for Molecular Neurology. The research at we have not obtained a subset of proteins which VIB-CMN is funded in part by the University of could reflect A + T+ profile. Further research in large, Antwerp Research Fund. FB is supported by the 1366 L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood NIHR biomedical research centre at UCLH. RD is (2019) ATN profiles among cognitively normal individuals supported by Health Data Research UK, the National and longitudinal cognitive outcomes. Neurology 92, e1567- e1579. Institute for Health Research (NIHR) Biomedical [6] De Ferrari GV, Chacon MA, Barria MI, Garrido JL, Godoy Research Centre at South London, Maudsley NHS JA, Olivares G, Reyes AE, Alvarez A, Bronfman M, Ine- Foundation Trust, King’s College London and the strosa NC (2003) Activation of Wnt signaling rescues NIHR University College London Hospitals Biomed- neurodegeneration and behavioral impairments induced by beta-amyloid fibrils. Mol Psychiatry 8, 195-208. ical Research Centre. LS is funded by DPUK through [7] Alvarez AR, Godoy JA, Mullendorff K, Olivares GH, MRC (grant no. MR/L023784/2) and the UK Med- Bronfman M, Inestrosa NC (2004) Wnt-3a overcomes beta- ical Research Council Award to the University of amyloid toxicity in rat hippocampal neurons. Exp Cell Res Oxford (grant no. MC PC 17215). Also support was 297, 186-196. [8] Vargas JY, Fuenzalida M, Inestrosa NC (2014) In vivo received from the NIHR Biomedical Research Centre activation of Wnt signaling pathway enhances cognitive at Oxford Health NHS Foundation Trust. function of adult mice and reverses cognitive deficits in an Authors’ disclosures available online (https:// Alzheimer’s disease model. J Neurosci 34, 2191-2202. www.j-alz.com/manuscript-disclosures/20-0208r3). [9] Toledo EM, Inestrosa NC (2010) Activation of Wnt signal- ing by lithium and rosiglitazone reduced spatial memory impairment and neurodegeneration in brains of an APP- DATA AVAILABILITY swe/PSEN1DeltaE9 mouse model of Alzheimer’s disease. Mol Psychiatry 15, 272-285, 228. [10] Tapia-Rojas C, Inestrosa NC (2018) Loss of canonical Wnt The data sets generated and analyzed during signaling is involved in the pathogenesis of Alzheimer’s the present study are available online. Briefly, disease. Neural Regen Res 13, 1705-1710. access to EMIF Somascan data could be obtained [11] Caricasole A, Copani A, Caraci F, Aronica E, Rozemuller via the EMIF-AD Catalogue (http://www.emif.eu/ AJ, Caruso A, Storto M, Gaviraghi G, Terstappen GC, Nicoletti F (2004) Induction of Dickkopf-1, a negative emif-catalogue/). Access to ANM Somascan data modulator of the Wnt pathway, is associated with neu- could be via AData (Viewer) (https://adata.scai. ronal degeneration in Alzheimer’s brain. J Neurosci 24, fraunhofer.de/). Research questions need to be sub- 6021-6027. mitted when applying for the access. In vitro [12] Rosi MC, Luccarini I, Grossi C, Fiorentini A, Spillan- tini MG, Prisco A, Scali C, Gianfriddo M, Caricasole A, Somascan data are available upon request. Terstappen GC, Casamenti F (2010) Increased Dickkopf-1 expression in transgenic mouse models of neurodegenera- SUPPLEMENTARY MATERIAL tive disease. J Neurochem 112, 1539-1551. [13] Bayod S, Felice P, Andres P, Rosa P, Camins A, Pallas M, Canudas AM (2015) Downregulation of canonical Wnt sig- The supplementary material is available in the naling in hippocampus of SAMP8 mice. Neurobiol Aging electronic version of this article: https://dx.doi. 36, 720-729. org/10.3233/JAD-200208. [14] Galli S, Lopes DM, Ammari R, Kopra J, Millar SE, Gibb A, Salinas PC (2014) Deficient Wnt signalling triggers stri- atal synaptic degeneration and impaired motor behaviour in REFERENCES adult mice. Nat Commun 5, 4992. [15] Sellers KJ, Elliott C, Jackson J, Ghosh A, Ribe E, Rojo AI, [1] Lane CA, Hardy J, Schott JM (2018) Alzheimer’s disease. Jarosz-Griffiths HH, Watson IA, Xia W, Semenov M, Morin Eur J Neurol 25, 59-70. P, Hooper NM, Porter R, Preston J, Al-Shawi R, Baillie G, [2] Jack CR Jr, Bennett DA, Blennow K, Carrillo MC, Dunn B, Lovestone S, Cuadrado A, Harte M, Simons P, Srivastava Haeberlein SB, Holtzman DM, Jagust W, Jessen F, Karlaw- DP, Killick R (2018) Amyloid beta synaptotoxicity is Wnt- ish J, Liu E, Molinuevo JL, Montine T, Phelps C, Rankin KP, PCP dependent and blocked by fasudil. Alzheimers Dement Rowe CC, Scheltens P, Siemers E, Snyder HM, Sperling R 14, 306-317. (2018) NIA-AA Research Framework: Toward a biological [16] Purro SA, Dickins EM, Salinas PC (2012) The secreted definition of Alzheimer’s disease. Alzheimers Dement 14, Wnt antagonist Dickkopf-1 is required for amyloid beta- 535-562. mediated synaptic loss. J Neurosci 32, 3492-3498. [3] Cummings J (2019) The National Institute on Aging- [17] Marzo A, Galli S, Lopes D, McLeod F, Podpolny M, Alzheimer’s Association Framework on Alzheimer’s Segovia-Roldan M, Ciani L, Purro S, Cacucci F, Gibb A, disease: Application to clinical trials. Alzheimers Dement Salinas PC (2016) Reversal of synapse degeneration by 15, 172-178. restoring Wnt signaling in the adult hippocampus. Curr Biol [4] Yu JT, Li JQ, Suckling J, Feng L, Pan A, Wang YJ, Song B, 26, 2551-2561. Zhu SL, Li DH, Wang HF, Tan CC, Dong Q, Tan L, Mok V, [18] Killick R, Ribe EM, Al-Shawi R, Malik B, Hooper C, Fer- Aisen PS, Weiner MM (2019) Frequency and longitudinal nandes C, Dobson R, Nolan PM, Lourdusamy A, Furney S, clinical outcomes of Alzheimer’s AT(N) biomarker profiles: Lin K, Breen G, Wroe R, To AW, Leroy K, Causevic M, A longitudinal study. Alzheimers Dement 15, 1208-1217. Usardi A, Robinson M, Noble W, Williamson R, Lunnon [5] Soldan A, Pettigrew C, Fagan AM, Schindler SE, Moghekar K, Kellie S, Reynolds CH, Bazenet C, Hodges A, Brion A, Fowler C, Li QX, Collins SJ, Carlsson C, Asthana S, JP, Stephenson J, Simons JP, Lovestone S (2014) Clus- Masters CL, Johnson S, Morris JC, Albert M, Gross AL terin regulates beta-amyloid toxicity via Dickkopf-1-driven L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood 1367 induction of the wnt-PCP-JNK pathway. Mol Psychiatry 19, J, Willemsen G, Boomsma DI, Yao J, Swords Jenny N, 88-98. Haritunians T, McKnight B, Lumley T, Taylor KD, Rot- [19] Lovestone S, Francis P, Kloszewska I, Mecocci P, Sim- ter JI, Psaty BM, Peters A, Gieger C, Illig T, Grotevendt mons A, Soininen H, Spenger C, Tsolaki M, Vellas B, A, Homuth G, Volzke H, Kocher T, Goel A, Franzosi Wahlund LO, Ward M, AddNeuroMed Consortium (2009) MG, Seedorf U, Clarke R, Steri M, Tarasov KV, Sanna AddNeuroMed–the European collaboration for the discov- S, Schlessinger D, Stott DJ, Sattar N, Buckley BM, Rum- ery of novel biomarkers for Alzheimer’s disease. Ann N Y ley A, Lowe GD, McArdle WL, Chen MH, Tofler GH, Acad Sci 1180, 36-46. Song J, Boerwinkle E, Folsom AR, Rose LM, Franco- [20] Bos I, Vos S, Vandenberghe R, Scheltens P, Engelborghs Cereceda A, Teichert M, Ikram MA, Mosley TH, Bevan S, Frisoni G, Molinuevo JL, Wallin A, Lleo A, Popp J, S, Dichgans M, Rothwell PM, Sudlow CL, Hopewell JC, Martinez-Lage P, Baird A, Dobson R, Legido-Quigley C, Chambers JC, Saleheen D, Kooner JS, Danesh J, Nelson Sleegers K, Van Broeckhoven C, Bertram L, Ten Kate M, CP, Erdmann J, Reilly MP, Kathiresan S, Schunkert H, Barkhof F, Zetterberg H, Lovestone S, Streffer J, Visser Morange PE, Ferrucci L, Eriksson JG, Jacobs D, Deary IJ, PJ (2018) The EMIF-AD Multimodal Biomarker Discov- Soranzo N, Witteman JC, de Geus EJ, Tracy RP, Hayward ery study: Design, methods and cohort characteristics. C, Koenig W, Cucca F, Jukema JW, Eriksson P, Seshadri Alzheimers Res Ther 10, 64. S, Markus HS, Watkins H, Samani NJ, Wallaschofski H, [21] Westman E, Simmons A, Muehlboeck JS, Mecocci P, Vel- Smith NL, Tregouet D, Ridker PM, Tang W, Strachan las B, Tsolaki M, Kloszewska I, Soininen H, Weiner MW, DP, Hamsten A, O’Donnell CJ (2013) Multiethnic meta- Lovestone S, Spenger C, Wahlund LO (2011) AddNeu- analysis of genome-wide association studies in>100 000 roMed and ADNI: Similar patterns of Alzheimer’s atrophy subjects identifies 23 fibrinogen-associated Loci but no and automated MRI classification accuracy in Europe and strong evidence of a causal association between circulat- North America. Neuroimage 58, 818-828. ing fibrinogen and cardiovascular disease. Circulation 128, [22] Hong S, Prokopenko D, Dobricic V, Kilpert F, Bos I, Vos 1310-1324. SJ, Tijms BM, Andreasson U, Blennow K, Vandenberghe [28] Lambert JC, Ibrahim-Verbaas CA, Harold D, Naj AC, Sims R (2019) Genome-wide association study of Alzheimer’s R, Bellenguez C, DeStafano AL, Bis JC, Beecham GW, disease CSF biomarkers in the EMIF-AD Multimodal Grenier-Boley B, Russo G, Thorton-Wells TA, Jones N, Biomarker Discovery dataset. bioRxiv, 774554. Smith AV, Chouraki V, Thomas C, Ikram MA, Zelenika D, [23] Gold L, Ayers D, Bertino J, Bock C, Bock A, Brody EN, Vardarajan BN, Kamatani Y, Lin CF, Gerrish A, Schmidt Carter J, Dalby AB, Eaton BE, Fitzwater T, Flather D, H, Kunkle B, Dunstan ML, Ruiz A, Bihoreau MT, Choi Forbes A, Foreman T, Fowler C, Gawande B, Goss M, Gunn SH, Reitz C, Pasquier F, Cruchaga C, Craig D, Amin N, M, Gupta S, Halladay D, Heil J, Heilig J, Hicke B, Husar Berr C, Lopez OL, De Jager PL, Deramecourt V, Johnston G, Janjic N, Jarvis T, Jennings S, Katilius E, Keeney TR, JA, Evans D, Lovestone S, Letenneur L, Moron FJ, Rubin- Kim N, Koch TH, Kraemer S, Kroiss L, Le N, Levine D, sztein DC, Eiriksdottir G, Sleegers K, Goate AM, Fievet N, Lindsey W, Lollo B, Mayfield W, Mehan M, Mehler R, Nel- Huentelman MW, Gill M, Brown K, Kamboh MI, Keller son SK, Nelson M, Nieuwlandt D, Nikrad M, Ochsner U, L, Barberger-Gateau P, McGuiness B, Larson EB, Green R, Ostroff RM, Otis M, Parker T, Pietrasiewicz S, Resnicow Myers AJ, Dufouil C, Todd S, Wallon D, Love S, Rogaeva DI, Rohloff J, Sanders G, Sattin S, Schneider D, Singer E, Gallacher J, St George-Hyslop P, Clarimon J, Lleo A, B, Stanton M, Sterkel A, Stewart A, Stratford S, Vaught Bayer A, Tsuang DW, Yu L, Tsolaki M, Bossu P, Spalletta JD, Vrkljan M, Walker JJ, Watrobka M, Waugh S, Weiss G, Proitsi P, Collinge J, Sorbi S, Sanchez-Garcia F, Fox NC, A, Wilcox SK, Wolfson A, Wolk SK, Zhang C, Zichi D Hardy J, Deniz Naranjo MC, Bosco P, Clarke R, Brayne C, (2010) Aptamer-based multiplexed proteomic technology Galimberti D, Mancuso M, Matthews F, Moebus S, Mecocci for biomarker discovery. PLoS One 5, e15004. P, Del Zompo M, Maier W, Hampel H, Pilotto A, Bullido [24] Grover S, Del Greco MF, Stein CM, Ziegler A (2017) M, Panza F, Caffarra P, Nacmias B, Gilbert JR, Mayhaus M, Mendelian randomization. Methods Mol Biol 1666, Lannefelt L, Hakonarson H, Pichler S, Carrasquillo MM, 581-628. Ingelsson M, Beekly D, Alvarez V, Zou F, Valladares O, [25] Davey Smith G, Hemani G (2014) Mendelian ran- Younkin SG, Coto E, Hamilton-Nelson KL, Gu W, Razquin domization: Genetic anchors for causal inference in C, Pastor P, Mateo I, Owen MJ, Faber KM, Jonsson PV, epidemiological studies. Hum Mol Genet 23, R89-98. Combarros O, O’Donovan MC, Cantwell LB, Soininen H, [26] Rasmussen KL, Nordestgaard BG, Frikke-Schmidt R, Blacker D, Mead S, Mosley TH, Jr., Bennett DA, Harris TB, Nielsen SF (2018) An updated Alzheimer hypothesis: Com- Fratiglioni L, Holmes C, de Bruijn RF, Passmore P, Montine plement C3 and risk of Alzheimer’s disease-A cohort study TJ, Bettens K, Rotter JI, Brice A, Morgan K, Foroud TM, of 95,442 individuals. Alzheimers Dement 14, 1589-1601. Kukull WA, Hannequin D, Powell JF, Nalls MA, Ritchie [27] Sabater-Lleal M, Huang J, Chasman D, Naitza S, Dehghan K, Lunetta KL, Kauwe JS, Boerwinkle E, Riemenschneider A, Johnson AD, Teumer A, Reiner AP, Folkersen L, Basu M, Boada M, Hiltuenen M, Martin ER, Schmidt R, Rujescu S, Rudnicka AR, Trompet S, Malarstig A, Baumert J, Bis D, Wang LS, Dartigues JF, Mayeux R, Tzourio C, Hofman JC, Guo X, Hottenga JJ, Shin SY, Lopez LM, Lahti J, A, Nothen MM, Graff C, Psaty BM, Jones L, Haines JL, Tanaka T, Yanek LR, Oudot-Mellakh T, Wilson JF, Navarro Holmans PA, Lathrop M, Pericak-Vance MA, Launer LJ, P, Huffman JE, Zemunik T, Redline S, Mehra R, Pulanic D, Farrer LA, van Duijn CM, Van Broeckhoven C, Moskv- Rudan I, Wright AF, Kolcic I, Polasek O, Wild SH, Camp- ina V, Seshadri S, Williams J, Schellenberg GD, Amouyel bell H, Curb JD, Wallace R, Liu S, Eaton CB, Becker DM, P (2013) Meta-analysis of 74,046 individuals identifies 11 Becker LC, Bandinelli S, Raikkonen K, Widen E, Palotie new susceptibility loci for Alzheimer’s disease. Nat Genet A, Fornage M, Green D, Gross M, Davies G, Harris SE, 45, 1452-1458. Liewald DC, Starr JM, Williams FM, Grant PJ, Spector [29] Marioni RE, Harris SE, Zhang Q, McRae AF, Hagenaars TD, Strawbridge RJ, Silveira A, Sennblad B, Rivadeneira SP, Hill WD, Davies G, Ritchie CW, Gale CR, Starr JM, F, Uitterlinden AG, Franco OH, Hofman A, van Dongen Goate AM, Porteous DJ, Yang J, Evans KL, Deary IJ, 1368 L. Shi et al. / Dickkopf-1 Overexpression in vitro Nominates Candidate Blood Wray NR, Visscher PM (2018) GWAS on family history [38] Ren C, Gu X, Li H, Lei S, Wang Z, Wang J, Yin P, Zhang of Alzheimer’s disease. Transl Psychiatry 8, 99. C, Wang F, Liu C (2019) The role of DKK1 in Alzheimer’s [30] Borenstein M, Hedges LV, Higgins JP, Rothstein HR (2010) disease: A potential intervention point of brain damage pre- A basic introduction to fixed-effect and random-effects vention? Pharmacol Res 144, 331-335. models for meta-analysis. Res Synth Methods 1, 97-111. [39] Anderton BH, Dayanandan R, Killick R, Lovestone S (2000) [31] Bowden J, Davey Smith G, Haycock PC, Burgess S (2016) Does dysregulation of the Notch and wingless/Wnt path- Consistent estimation in Mendelian randomization with ways underlie the pathogenesis of Alzheimer’s disease? Mol some invalid instruments using a weighted median estima- Med Today 6, 54-59. tor. Genet Epidemiol 40, 304-314. [40] Libro R, Bramanti P, Mazzon E (2016) The role of the Wnt [32] Bowden J, Davey Smith G, Burgess S (2015) Mendelian canonical signaling in neurodegenerative diseases. Life Sci randomization with invalid instruments: Effect estimation 158, 78-88. and bias detection through Egger regression. Int J Epidemiol [41] Inestrosa NC, Varela-Nallar L (2014) Wnt signaling in the 44, 512-525. nervous system and in Alzheimer’s disease. J Mol Cell Biol [33] Hemani G, Zheng J, Wade KH, Laurin C, Elsworth B, 6, 64-74. Burgess S, Bowden J, Langdon R, Tan V, Yarmolin- [42] Kucukkilic E, Brookes K, Barber I, Guetta-Baranes T, Mor- sky J (2016) MR-Base: A platform for systematic causal gan K, Hollox EJ (2018) Complement receptor 1 gene (CR1) inference across the phenome using billions of genetic asso- intragenic duplication and risk of Alzheimer’s disease. Hum ciations. BioRxiv, 078972. Genet 137, 305-314. [34] Shi L, Baird AL, Westwood S, Hye A, Dobson R, [43] Counts SE, Ikonomovic MD, Mercado N, Vega IE, Mufson Thambisetty M, Lovestone S (2018) A decade of blood EJ (2017) Biomarkers for the early detection and progres- biomarkers for Alzheimer’s disease research: An evolving sion of Alzheimer’s disease. Neurotherapeutics 14, 35-53. field, improving study designs, and the challenge of repli- [44] Keshavan A, Heslegrave A, Zetterberg H, Schott JM (2017) cation. J Alzheimers Dis 62, 1181-1198. Blood biomarkers for Alzheimer’s disease: Much promise, [35] Baird AL, Westwood S, Lovestone S (2015) Blood-based cautious progress. Mol Diagn Ther 21, 13-22. proteomic biomarkers of Alzheimer’s disease pathology. [45] Wyss-Coray T, Yan F, Lin AH, Lambris JD, Alexander JJ, Front Neurol 6, 236. Quigg RJ, Masliah E (2002) Prominent neurodegeneration [36] Ashton NJ, Kiddle SJ, Graf J, Ward M, Baird AL, Hye A, and increased plaque formation in complement-inhibited Westwood S, Wong KV, Dobson RJ, Rabinovici GD, Miller Alzheimer’s mice. Proc Natl Acad Sci U S A 99, 10837- BL, Rosen HJ, Torres A, Zhang Z, Thurfjell L, Covin A, 10842. Hehir CT, Baker D, Bazenet C, Lovestone S (2015) Blood [46] Maier M, Peng Y, Jiang L, Seabrook TJ, Carroll MC, Lemere protein predictors of brain amyloid for enrichment in clinical CA (2008) Complement C3 deficiency leads to acceler- trials? Alzheimers Dement (Amst) 1, 48-60. ated amyloid beta plaque deposition and neurodegeneration [37] Palomer E, Buechler J, Salinas PC (2019) Wnt Signaling and modulation of the microglia/macrophage phenotype in deregulation in the aging and Alzheimer’s brain. Front Cell amyloid precursor protein transgenic mice. J Neurosci 28, Neurosci 13, 227. 6333-6341.