doi:10.1093/brain/awaa325 BRAIN 2020: 143; 3776–3792 | 3776 Pathophysiological subtypes of Alzheimer’s disease based on cerebrospinal fluid proteomics Betty M. Tijms,1 Johan Gobom,2,3 Lianne Reus,1 Iris Jansen,1 Shengjun Hong,4 Valerija Dobricic,4 Fabian Kilpert,4 Mara ten Kate,1 Frederik Barkhof,5,6 Magda Tsolaki,7 Frans R. J. Verhey,8 Julius Popp,9,10 Pablo Martinez-Lage,11 Rik Vandenberghe,12,13 Alberto Lleó,14 José Luı́s Molinuevo,15,16 Sebastiaan Engelborghs,17,18 Lars Bertram,4 Simon Lovestone,19,20 Johannes Streffer,17,21 Stephanie Vos,8 Isabelle Bos,1,8 The Alzheimer’s Disease Neuroimaging Initiative (ADNI),† Kaj Blennow,2,3 Philip Scheltens,1 Charlotte E. Teunissen,22 Henrik Zetterberg2,3,23,24 and Pieter Jelle Visser1,8,25 † Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf Alzheimer’s disease is biologically heterogeneous, and detailed understanding of the processes involved in patients is critical for de- velopment of treatments. CSF contains hundreds of proteins, with concentrations reflecting ongoing (patho)physiological processes. This provides the opportunity to study many biological processes at the same time in patients. We studied whether Alzheimer’s dis- ease biological subtypes can be detected in CSF proteomics using the dual clustering technique non-negative matrix factorization. In two independent cohorts (EMIF-AD MBD and ADNI) we found that 705 (77% of 911 tested) proteins differed between Alzheimer’s disease (defined as having abnormal amyloid, n = 425) and controls (defined as having normal CSF amyloid and tau and normal cognition, n = 127). Using these proteins for data-driven clustering, we identified three robust pathophysiological Alzheimer’s disease subtypes within each cohort showing (i) hyperplasticity and increased BACE1 levels; (ii) innate immune activa- tion; and (iii) blood–brain barrier dysfunction with low BACE1 levels. In both cohorts, the majority of individuals were labelled as having subtype 1 (80, 36% in EMIF-AD MBD; 117, 59% in ADNI), 71 (32%) in EMIF-AD MBD and 41 (21%) in ADNI were labelled as subtype 2, and 72 (32%) in EMIF-AD MBD and 39 (20%) individuals in ADNI were labelled as subtype 3. Genetic analyses showed that all subtypes had an excess of genetic risk for Alzheimer’s disease (all P 4 0.01). Additional pathological com- parisons that were available for a subset in ADNI suggested that subtypes showed similar severity of Alzheimer’s disease path- ology, and did not differ in the frequencies of co-pathologies, providing further support that found subtypes truly reflect Alzheimer’s disease heterogeneity. Compared to controls, all non-demented Alzheimer’s disease individuals had increased risk of showing clinical progression (all P 5 0.01). Compared to subtype 1, subtype 2 showed faster clinical progression after correcting for age, sex, level of education and tau levels (hazard ratio = 2.5; 95% confidence interval = 1.2, 5.1; P = 0.01), and subtype 3 at trend level (hazard ratio = 2.1; 95% confidence interval = 1.0, 4.4; P = 0.06). Together, these results demonstrate the value of CSF proteomics in studying the biological heterogeneity in Alzheimer’s disease patients, and suggest that subtypes may require tailored therapy. 1 Alzheimer Center Amsterdam, Department of Neurology, Amsterdam Neuroscience, Vrije Universiteit Amsterdam, Amsterdam UMC - Location VUmc, The Netherlands Received March 23, 2020. Revised August 03, 2020. Accepted August 06, 2020. Advance access publication November 18, 2020 C The Author(s) (2020). Published by Oxford University Press on behalf of the Guarantors of Brain. V This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@oup.com Pathophysiological Alzheimer’s disease subtypes BRAIN 2020: 143; 3776–3792 | 3777 2 Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden 3 Department of Psychiatry and Neurochemistry, Institute of Neuroscience and Physiology, Sahlgrenska Academy at the University of Gothenburg, Mölndal, Sweden 4 Lübeck Interdisciplinary Platform for Genome Analytics (LIGA), Institutes of Neurogenetics and Cardiogenetics, University of Lübeck, Lübeck, Germany 5 Department of Radiology and Nuclear Medicine, Amsterdam Neuroscience, Vrije Universiteit Amsterdam, Amsterdam UMC - loca- tion VUmc, Amsterdam, The Netherlands 6 Institutes of Neurology and Healthcare Engineering, UCL London, London, UK 7 1st Department of Neurology, AHEPA University Hospital, Makedonia, Thessaloniki, Greece 8 Alzheimer Center Limburg, School for Mental Health and Neuroscience, Maastricht University, Maastricht, The Netherlands 9 University Hospital Lausanne, Lausanne, Switzerland 10 Geriatric Psychiatry, Department of Psychiatry, Geneva University Hospital, Geneva, Switzerland 11 Fundación CITA-Alzhéimer Fundazioa, San Sebastian, Spain 12 Neurology Service, University Hospitals Leuven, Leuven, Belgium 13 Laboratory for Cognitive Neurology, Department of Neurosciences, KU Leuven, Leuven, Belgium 14 IIB-Sant Pau, Hospital de la Santa Creu i Sant Pau, Universitat Autonoma de Barcelona, Barcelona, Spain 15 Barcelonabeta Brain Research Center (BBRC), Pasqual Maragall Foundation, Barcelona, Spain 16 Alzheimer’s Disease Unit and Other Cognitive Disorders Unit, Hospital Clinic de Barcelona, Barcelona, Spain 17 Institute Born-Bunge, Reference Center for Biological Markers of Dementia (BIODEM), Institute Born-Bunge, University of Antwerp, Belgium 18 Department of Neurology, UZ Brussel and Center for Neurosciences (C4N), Vrije Universiteit Brussel (VUB), Brussels, Belgium 19 University of Oxford, Oxford, UK 20 Janssen R&D, Beerse, Belgium 21 UCB Biopharma SPRL, Brain-l’Alleud, Belgium 22 Neurochemistry laboratory, Department of Clinical Chemistry, Amsterdam UMC - location VUmc, Amsterdam Neuroscience, The Netherlands 23 Department of Neurodegenerative Disease, UCL Institute of Neurology, London, UK 24 UK Dementia Research Institute at UCL, London, UK 25 Department of Neurobiology, Care Sciences and Society, Division of Neurogeriatrics, Karolinska Institutet, Stockholm Sweden Correspondence to: B. M. Tijms, PhD Alzheimer Center and Department of Neurology, Amsterdam UMC location VUmc, Amsterdam Neuroscience Amsterdam PO Box 7057, 1007 MB, Amsterdam The Netherlands E-mail: b.tijms@amsterdamumc.nl Keywords: Alzheimer’s disease; cerebrospinal fluid; proteomics; subtypes Abbreviations: ADNI = Alzheimer’s Disease Neuroimaging Initiative; EMIF-AD MBD = European Medical Information Framework for Alzheimer’s Disease Multimodal Biomarker Discovery; NMF = non-negative matrix factorization; MCI = mild cog- nitive impairment; SNP = single nucleotide polymorphism 2015; Hondius et al., 2016). Part of the heterogeneity in Introduction Alzheimer’s disease is explained by genetic variance (Ridge Alzheimer’s disease is a neurodegenerative disorder and the et al., 2016) indicating that multiple biological pathways are most common cause of dementia. The pathological hall- involved in Alzheimer’s disease, and these include processes marks are amyloid plaques and tau neurofibrillary tangles in related to amyloid and tau metabolism, the innate immune the brain. Biomarkers for amyloid and tau pathology are system, lipid processing, and synaptic functioning (Lambert therefore part of the biological definition of Alzheimer’s dis- et al., 2013; Kunkle et al., 2019; Jansen et al., 2020). It is ease (Dubois et al., 2007, 2014; Albert et al., 2011; Jack likely that patients will require personalized medicine de- et al., 2011, 2018; Sperling et al., 2011). The current defin- pending on the molecular processes involved, but at this ition implies that Alzheimer’s disease is a single disease en- point there are no tools to identify biological subtypes in tity. However, individuals with Alzheimer’s disease show Alzheimer’s disease in vivo. CSF contains many proteins that considerable variability in terms of clinical symptoms, age of reflect (patho)physiological processes in the brain, and could onset, disease progression, cortical atrophy patterns, CSF provide insight into biological processes involved in levels of tau, and other pathological markers (Blennow and Alzheimer’s disease. Wallin, 1992; Iqbal et al., 2005; van der Vlies et al., 2009; Previous studies examining heterogeneity in Alzheimer’s Wallin et al., 2010; Whitwell et al., 2012; Lam et al., 2013; disease based on CSF levels of targeted proteins amyloid, Möller et al., 2013; Ossenkoppele et al., 2015; Smits et al., tau and p-tau and/or ubiquitin, suggest that at least three 3778 | BRAIN 2020: 143; 3776–3792 B. M. Tijms et al. subtypes exist, mainly characterized by having low, inter- mediate or high tau levels (Iqbal et al., 2005; van der Vlies Materials and methods et al., 2009; Wallin et al., 2010). Unbiased or large-scale tar- geted proteomic CSF analyses have potential to further re- Participants fine which biological processes become disrupted in We selected individuals with CSF amyloid-b1–42, tau, and pro- Alzheimer’s disease. So far, Alzheimer’s disease proteomic teomics data from two independent multicentre Alzheimer’s dis- studies mostly focussed on finding novel biomarkers by com- ease studies, the EMIF-AD MBD (Bos et al., 2018) and the paring patients with Alzheimer’s disease with control sub- ADNI (adni.loni.usc.edu). Both cohorts included individuals jects (Maarouf et al., 2009; Meyer et al., 2018; for reviews with normal cognition, mild cognitive impairment (MCI) or Alzheimer’s disease-type dementia, as determined according to see Pedrero-Prieto et al., 2020; Wesenhagen et al., 2020), international consensus criteria (McKhann et al., 1984, 2011; and so it remains unclear whether pathophysiological sub- Petersen et al., 1999; Winblad et al., 2004). Control was defined types within Alzheimer’s disease can be discovering with by normal cognition and normal CSF amyloid-b1–42 and tau CSF proteomics. Furthermore, if genetic variance in biomarkers (see below), and Alzheimer’s disease pathological Alzheimer’s disease risk genes contributes to interindividual change was defined by abnormal CSF amyloid-b1–42 (Jack et al., variability in underlying disease mechanisms, it can be 2018). Both studies excluded patients with any neurological dis- hypothesized that these should already be detectable in pre- ease other than suspected Alzheimer’s disease, such as symptomatic stages of Alzheimer’s disease. Parkinson’s disease, dementia with Lewy bodies, frontotemporal In this study we used a data-driven dual clustering tech- dementia, progressive supranuclear palsy, corticobasal syn- nique to identify biological subtypes of Alzheimer’s disease drome, normal pressure hydrocephalus, or vascular dementia. ADNI began in 2003 as a public-private collaboration under in CSF proteomics in two large independent Alzheimer’s dis- the supervision of principle investigator Michael W. Weiner, ease cohorts [i.e. the European Medical Information MD. The primary goal of ADNI is to study whether serial MRI, Framework for Alzheimer’s Disease Multimodal Biomarker PET, other biological markers, and clinical and neuropsycho- Discovery (EMIF-AD MBD) and the Alzheimer’s Disease logical measures can be combined to measure the progression of Neuroimaging Initiative (ADNI)] across the clinical spec- MCI and early Alzheimer’s disease. See www.adni-info.org for trum. We defined Alzheimer’s disease by the presence of the latest information. The institutional review boards of all par- amyloid pathology as indicated by abnormal levels of CSF ticipating institutions approved the procedures for this study. amyloid-b1–42, because abnormal amyloid-b1–42 CSF shows Written informed consent was obtained from all participants or high concordance with the presence of amyloid and tau surrogates. pathology upon neuropathological examination (Shaw et al., 2009). In contrast, CSF tau levels show more variability CSF data amongst patients, with up to 30% of individuals with patho- CSF samples were obtained as previously described (Shaw, logically confirmed Alzheimer’s disease showing normal lev- et al., 2009; Toledo et al., 2013; Bos et al., 2018). CSF amyloid- els of CSF tau (Shaw et al., 2009). Therefore, we used CSF b1–42, t-tau and p-tau levels were measured with INNOTEST total (t)-tau and phosphorylated (p)-tau as independent out- ELISAsor INNOBIA AlzBio3 kit (Fujirebio) in EMIF-AD MBD come markers. We further excluded patients that had evi- (Bos et al., 2018), and in ADNI with the multiplex xMAP lumi- dence of known neurodegenerative disorders associated with nex platform (Luminex Corp) at the ADNI Biomarker Core la- amyloid aggregation other than Alzheimer’s disease. We first boratory at the University of Pennsylvania Medical Center. For identified which proteins were associated with Alzheimer’s the ADNI cohort, biomarker abnormality was defined by amyl- disease. Next, we used unsupervised clustering on these pro- oid-b1–42 levels 5192 pg/ml, and t-tau levels 493 pg/ml (Shaw et al., 2009). In the EMIF-AD MBD cohort, cut-offs for p-tau teins to identify biological subtypes of Alzheimer’s disease. and t-tau were study-specific, as previously reported (Bos et al., We interpreted Alzheimer’s disease subtype protein profiles 2018). For amyloid-b1–42 cut-offs, the studies in EMIF-AD in terms of biological processes through enrichment analyses, MBD differed in methodologies used to determine cut-offs, and performed post hoc analyses to characterize Alzheimer’s which may lead to bias (Bertens et al., 2017). To minimize such disease subtypes in terms of: clinical and biological charac- bias across studies, we determined centre-specific cut-offs using teristics known to be associated with Alzheimer’s disease i.e. unbiased Gaussian mixture modelling (Supplementary Table 1) established CSF markers (neurogranin, BACE1 activity, neu- (De Meyer, 2010; Bertens et al., 2017; Tijms et al., 2018). rofilament light, VILIP, YKL-40, sTREM2), APOE geno- Cluster analyses were performed on proteomic data using the type, Alzheimer’s disease polygenic risk scores, MRI markers tandem mass tag (TMT) technique with 10 + 1 plexing in for cortical atrophy, cognitive functioning and decline. EMIF-AD MBD using high pH reverse phase HPLC for peptide prefractionation (Batth et al., 2014; Magdalinou et al., 2017) Furthermore, we compared subtypes on vascular comorbid- (Supplementary material). For EMIF-AD MBD, the median ity using MRI markers for vascular damage. Finally, we (interquartile range, IQR) analytical coefficient of variation compared subtypes on neuropathological measures that across included proteins was 5.6 (3.8, 8.0) (see Supplementary were available for a subset of individuals (ADNI only), and Table 2 for protein-specific coefficients of variation). The EMIF- we assessed stability of proteomic subtypes over time for a AD MBD mass spectrometry proteomics data have been depos- subset of individuals who had longitudinal proteomics avail- ited to the ProteomeXchange Consortium via the PRIDE (Perez- able (ADNI only). Riverol et al., 2018) partner repository with the dataset Pathophysiological Alzheimer’s disease subtypes BRAIN 2020: 143; 3776–3792 | 3779 identifiers PXD019910 and 10.6019/PXD019910. Normalized prior to data analyses if they had a minor allele frequency abundances with associated clinical data can be requested from 52%, deviated significantly from Hardy-Weinberg equilibrium the EMIF-AD MBD consortium (Bos et al., 2018). In the ADNI (P 5 1  10–6) in the total sample of founder individuals, or cohort, four proteins included were determined with ELISAs, had a call rate of 598%. We only used SNPs with 55% geno- 311 protein fragments determined with Multi Reaction type missingness and removed samples with excess heterozygos- Monitoring (MRM) targeted mass spectroscopy, and 83 pro- ity rate (45 SD). After filtering, the genotype data in ADNI teins measured with the Rules Based Medicine (RBM) multiplex. included 1 496 949 SNPs and 6 706 731 SNPs in EMIF-AD Information on protein assessment and quality control is MBD. To control for population stratification, five principal described at http://adni.loni.usc.edu/data-samples/biospecimen- components were computed on a subset of relatively uncorre- data/. For ADNI MRM we used the quality-controlled finalized lated (r2 5 0.2) SNPs (PC1–PC5). Polygenic risk scores for ‘normalized intensity’ data (Spellman et al., 2015) (for detailed Alzheimer’s disease were calculated by adding the sum of each explanation of the normalization procedure see the ‘Biomarkers allele weighted by the strength of its association with Consortium CSF Proteomics MRM dataset’ in the ‘Data Primer’ Alzheimer’s disease risk using PRSice (Euesden et al., 2014). document at adni.loni.ucla.edu). All protein (EMIF-AD MBD The strength of these associations was calculated previously by and ADNI) and protein fragment (ADNI) values were first nor- the International Genomics of Alzheimer’s project (IGAP) gen- malized according to mean and standard deviation (SD) values ome-wide association study (GWAS) (Lambert et al., 2013). of the control group. Then, for ADNI, protein fragments from Clumping was performed prior to calculating polygenic risk MRM measurements were combined into a protein score when score to remove SNPs that are in linkage disequilibrium (r2 these correlated with r 4 0.5, and fragments that did not correl- 50.1) within a slicing 1 Mbp window. After clumping we com- ate were omitted from the present analyses. Eleven proteins puted 14 PGRSs with varying SNP inclusion threshold (P 5 10–30 were measured by different platforms in ADNI, for which values to P 5 0.5). Finally, we constructed specialized polygenic risk were averaged if they correlated with r 4 0.5, or we selected the scores including only SNPs that corresponded to genes part of the protein as measured by MRM [mean r = 0.74; min r = –0.50, GO pathways ‘innate immune response’ and ‘complement activa- max r = 0.92; for one protein RBM was excluded, another pro- tion’ for SNP inclusion thresholds (P 5 10–30 to P 5 1). All poly- tein (CST3) showed a strong anticorrelation between RBM and genic risk scores were regressed on PC1–PC3. MRM of r = –0.85, and was excluded]. Only proteins that were observed in 100% of the sample were considered for sub- sequent analyses, resulting in a total 707 proteins in EMIF-AD Cluster analyses with non-negative MBD and 204 proteins in ADNI. A subset of individuals had matrix factorization additional protein measurements available, which we excluded from clustering to use as independent outcomes for subtype in- First, in each cohort we selected proteins for clustering that dif- terpretation. In ADNI these were amyloid-b1–40 and amyloid- fered between the control and Alzheimer’s disease groups at un- b1–38, measured with 2D-UPLC tandem mass spectrometry, corrected P 5 0.10 using Kruskal-Wallis tests. As protein levels BACE1 activity, and ELISA measures of neurogranin, neurofila- can change non-linearly with levels of neuronal injury and/or ment light, VILIP, YKL40, SNAP25 and sTREM2. In EMIF-AD disease severity (De Leon et al., 2018; Duits et al., 2018), we MBD measurements were available for amyloid-b1–40, amyloid- repeated analyses stratifying Alzheimer’s disease individuals on b1–38, neurogranin, neurofilament light, and YKL-40 (Bos et al., disease stage (i.e. normal cognition, MCI and dementia), and on 2018). A subset of 70 (29%) ADNI individuals had repeated the presence of abnormal CSF levels of the neuronal injury MRM for 62 proteins [median five repeated measures; over me- marker t-tau. Next, we clustered these proteins with non-nega- dian (IQR) of 6 (4.3, 6.7) years for cognitively normal subjects tive matrix factorization (NMF). NMF is a dual clustering ap- and median (IQR) of 4 (3.9, 6.0) years for Alzheimer’s disease proach that is based on decomposition of the data by parts, patients], which we used to study the stability of proteomic sub- which reduces the dimensionality of data protein expression lev- types. For these analyses, we first standardized the proteins lev- els into fewer components that we consider protein profiles (Lee els according to the baseline mean and SD levels of the control and Seung, 1999), and at the same time this algorithm groups group, and then we constructed proteomic profile scores by subjects together into subtypes based on how well their protein averaging levels of proteins specific for a subtype. expression levels match the protein profiles. A strength of NMF compared to correlation-based approaches is that it is able cap- ture non-linear patterns associated with a certain subtype. To Genetic analyses aid interpretation of the results, we labelled proteins according ADNI samples were genotyped using either the Illumina 2.5-M to which subtype showed the highest average levels. We used array (a byproduct of the ADNI whole-genome sequencing sam- the R package NMF for clustering, with the ‘nonsmooth’ option ple) or the Illumina OmniQuad array (Saykin et al., 2010). that ensures sparse cluster solutions with enhanced separability APOE genotype was assessed with two single nucleotide poly- (Gaujoux and Seoighe, 2010). The NMF algorithm is stochastic morphisms (SNPs) (rs429358, rs7412) that define the e2, 3, and and therefore subject classification to a subtype can vary from 4 alleles, using DNA extracted by Cogenics from a 3 ml aliquot run to run, based on the random initial conditions. We assessed of EDTA blood. EMIF-AD MBD samples were genotoyped at stability of subtype classification over 50 different runs of NMF the USKH site using the Global Screening Array (Illumina, Inc) with the co-phonetic coefficient, with values ranging from 0 (i.e. (see Hong et al., 2019 for more details on imputation prepro- unstable solution) to 1 (i.e. subjects are always classified the cessing). In the ADNI cohort, SNPs were imputed using the same). We tested up to five clusters, and the optimal number of 1000 Genomes reference panel, with the use of the Michigan clusters was determined as the number of clusters for which: (i) imputation server. Genotype data were quality checked for gen- the co-phonetic correlation was high; (ii) fit compared to a der mismatch, relatedness and ancestry. SNPs were excluded lower cluster number solution was improved at least 2-fold over 3780 | BRAIN 2020: 143; 3776–3792 B. M. Tijms et al. a random solution; and (iii) silhouette width of the cluster solu- tion was 4 0.5. Clustering analyses were performed separately Results for each cohort. We performed pathway enrichment analysis for We included 127 control subjects with normal cognition and proteins that were characteristic for each subtype using the on- normal CSF amyloid-b1–42 and tau, and 425 individuals line Panther application (Mi et al., 2013). We used the with Alzheimer’s disease across the clinical spectrum [89 ENCODE and ChEA consensus transcription factor database in (21%) normal cognition, 195 (46%) MCI, and 141 (33%) the Enrichr webserver to identify potential upstream drivers of Alzheimer’s disease-type dementia]. Compared to controls, subtype-specific protein alterations (Chen et al., 2013; Kuleshov individuals with Alzheimer’s disease more often carried an et al., 2016). We selected pathways that were most consistently APOE e4 allele, had lower MMSE scores, and more often associated with the subtypes for visualization, and report all abnormal CSF p-tau and t-tau in both cohorts (Table 1). observed pathways in the Supplementary material. To determine Other characteristics were similar between groups in both cell type production we used the BRAIN RNASeq database (http://www.brainrnaseq.org) (Zhang et al., 2014). Proteins cohorts, except that individuals with Alzheimer’s disease were labelled as being specifically produced by a certain cell were older than controls in the EMIF-AD MBD. Relative to type when levels were 450% of the total produced across cell controls, individuals with Alzheimer’s disease showed differ- types, as non-specific when none of the cell types was 450%, ential CSF levels for 556 of 707 proteins (79%) measured in or as not detected when levels were all 50.2. EMIF-AD MBD and 149 of 204 (73%) proteins measured in ADNI (Supplementary Table 2). These Alzheimer’s dis- ease-specific proteins were considered for cluster analyses Statistical procedures for post hoc with NMF within in each cohort. subtype comparisons We performed the following post hoc comparisons of subtypes: Three biological Alzheimer’s disease CSF levels of t-tau, p-tau and other established Alzheimer’s dis- subtypes detected in CSF ease CSF markers that were not included in the cluster analyses to provide further independent interpretation of the cluster solu- proteomic data tions, age, gender, disease stage, APOE e4 genotype, According to our fit criteria, three clusters best described the Alzheimer’s disease PGRS, pathological measures, cortical thick- CSF proteomic data in both cohorts (Supplementary Table ness measures from 34 cortical areas as defined by the Desikan- 3). Repeating clustering of proteins using a Louvain modu- Killiany atlas (averaged over the left and right hemispheres) (see larity algorithm on a weighted protein co-expression net- Bos et al., 2018 for EMIF-AD MBD and http://adni.loni.cule. work also resulted in three protein clusters, which showed edu/ for detailed documentation on variable specific methods in good correspondence with the NMF protein clusters of 80% ADNI), vascular damage (visual ratings in EMIF-AD MBD, and in EMIF-AD MBD and 86% in ADNI (Supplementary white matter hyperintensity volumes in ADNI), Mini-Mental State Examination (MMSE) scores, level of education, neuro- Tables 5A and B). A 3D plot of subject loadings on clusters psychological test scores covering the memory (memory immedi- revealed a subset of five individuals with extreme loadings in ate and delayed recall scores on the logical memory subscale II EMIF-AD MBD (Supplementary Fig. 1). These individuals of the Wechsler Memory Scale), language (Boston Naming Test, did not show differences with other Alzheimer’s disease indi- and animal fluency), visuospatial processing (Clock drawing) viduals in terms of sample characteristics (Supplementary and attention/executive domains [digit span, Trail Making Test Table 4). To avoid potential overfitting, we repeated cluster (TMT) A and TMT B]. All continuous variables (except for age, analyses excluding these individuals, and a three-cluster solu- MMSE, and years of education) were standardized according to tion remained most optimal. We next labelled individuals the mean and SD of the control group. Subtype comparisons according to the subtype they scored highest on (Fig. 1A). In were performed with general linear models in case of continuous both cohorts, the majority of individuals were labelled as variables with two-sided testing, and with Chi-square tests for having subtype 1 [80 (36%) in EMIF-AD MBD, 117 (59%) discrete variables. Comparisons for continuous variables were in ADNI]; 71 (32%) in EMIF-AD MBD and 41 (21%) in performed without and with adjustment for age and sex, and ADNI were labelled as subtype 2; and 72 (32%) in EMIF- cognitive measures were additionally adjusted for level of educa- AD MBD and 39 (20%) individuals in ADNI were labelled tion. We used the R package ‘emmeans’ to obtain estimated as subtype 3. A subset of 92 proteins was measured in both marginalized means. Finally, we performed an exploratory ran- EMIF-AD MBD and ADNI, which showed consistent sub- dom forest analysis on the subset of 92 proteins that were asso- type differences in levels for 84–98% of proteins across the ciated with Alzheimer’s disease in both EMIF-AD MBD and ADNI to identify the best subset of proteins that can separate cohorts (Supplementary Fig. 2 and Supplementary Table the subtypes in the pooled sample and for each cohort separate- 5A). An exploratory random forest analysis on this subset of ly. For this, we used a bootstrap approach that fitted the model 92 overlapping proteins identified a subset of five proteins in two-thirds of the dataset and tested classification performance (SLITRK1, NEO1, EXTL2, TNFRSF21, IGFBP2) that best in the left out one-third, repeated 10 000 times in order to esti- separate the subtypes, with an overall accuracy of 80% mate 95% confidence intervals (CI) for classification performan- (95% CI: 74–86%) in the pooled sample, and similar per- ces. ADNI data were downloaded on 30 March 2018. All formance within cohorts (EMIF: 80%, 95% CI: 71%–88%; analyses were performed in R v3.5.1 ‘Feather Spray’. ADNI: 80%, 95% CI: 71%–89%), further supporting that Pathophysiological Alzheimer’s disease subtypes BRAIN 2020: 143; 3776–3792 | 3781 Table 1 Participant descriptions Descriptive EMIF-AD MBD ADNI Controls (n = 82) Alzheimer’s disease Controls (n = 45) Alzheimer’s disease (n = 228) (n = 197) Cognitive status, n (%) Intact cognition 82 (100) 57 (25) 45 (100) 32 (16) MCI 0 (0) 92 (40)* 0 (0) 103 (52) Dementia 0 (0) 79 (35) 0 (0) 62 (31) Age in years, mean (SD) 61.1 (7)** 68.1 (8)*,*** 75.8 (6) 74.9 (7) Female, n (%) 47 (57) 126 (55)* 23 (51) 81 (41) Years of education, mean (SD) 11.9 (3.5)** 11.2 (3.5)* 15.6 (3) 15.6 (3) MMSE, mean (SD)a 28.6 (1.3)** 25.6 (3.9)*** 29.2 (0.6) 26.1 (2.6)*** APOE e4, at least one allele (%)b 14 (22)** 140 (52) 4 (8) 129 (65)*** Hippocampal volume, mean (SD)c,e 0 (1) –1.4 (1.5)*,*** 0 (1) –1.7 (1.4)*** Amyloid-b1–42 pg/ml, mean (SD)f 0 (1) –2.8 (1.5)*,*** 247.5 (29.2) 139.1 (23.1)*** t-tau pg/ml, mean (SD)f 0 (1) 4.4 (4.7)*** 57.1 (13.1) 114.3 (54.9)*** Abnormal t-tau, n (%)g 0 (0) 151 (66)*** 0 (0) 115 (58)*** p-tau pg/ml, mean (SD)d,f 0 (1) 2.1 (2.5)*** 20.3 (9.4) 39.1 (17.5)*** Abnormal p-tau, n (%)g 7 (8.5) 149 (65)*,*** 9 (20) 168 (85)*** MMSE = Mini-Mental State Examination. a Data missing for one individual; b16 individuals; c159 individuals; and dfive individuals. e Scaled according to the mean and SD values in controls. f Based on Luminex in ADNI and scaled for EMIF according to cohort-specific controls as previously described (Bos et al., 2018). g Cut-off points to define abnormal levels for ADNI: t-tau 4 93 pg/ml, p-tau 4 21 pg/ml (Shaw et al., 2009), and cohort specific for EMIF-AD MBD as previously described (Bos et al., 2018). Groups were compared with v2 tests or t-test where appropriate. *P 5 0.05 for across cohort comparisons between Alzheimer’s disease groups. **P 5 0.05 for across cohort comparisons between controls. ***P 5 0.01  10–30 for controls versus Alzheimer’s disease within cohort. Cohort differences in continuous amyloid-b1-42, t-tau and p-tau levels are based on cohort scaled values according to control group levels. subtype definitions are robust. Individuals with subtype 1, (75%) in EMIF-AD MBD; 120 (81%) in ADNI; P-values compared to controls, had significantly higher levels for the ranging between 0.02  10–22 and 0.049] that mirrored the majority of proteins in both cohorts [EMIF-AD MBD: 309 increases observed in subtype 1, which suggests that type 3 (56%); ADNI: 92 (65%); P-values ranging between has neuronal hypoplasticity. Another group of proteins was 0.02  10–21 and 0.049] (Fig. 1B and Supplementary Table specifically increased in subtype 3 compared to control sub- 5A and B). The predominant cell types producing these pro- jects [76 (14%) in EMIF-AD MBD; 6 (4%) in ADNI], teins were neurons and astrocytes in both cohorts (Fig. 1C including albumin and immunoglobulin proteins, of which and Supplementary Table 6). GO pathway analyses for pro- higher CSF levels have been reported with blood–brain bar- teins increased in subtype 1 showed in both EMIF-AD MBD rier dysfunction (Dayon et al., 2019). GO pathway analyses and ADNI enrichment for processes related to the MAPK/ for proteins specifically increased in subtype 3 were per- ERK cascade, synaptic structure and function, axonal devel- formed for EMIF-AD MBD only, as the number of subtype opment, and glucose metabolism, suggesting that subtype 1 3-specific increased proteins in ADNI was too low (note that shows neuronal hyperplasticity (Fig. 1D and Supplementary for proteins showing subtype 3-specific decreases 99% of Table 7) (77% processes enriched in ADNI overlapped with the pathways enriched in ADNI overlapped with EMIF-AD those enriched in EMIF-AD MBD). Subtype 2 also showed MBD, and these pathways involved 66% of those enriched mostly higher protein levels than controls [EMIF-AD MBD: for subtype 1 increased proteins). Subtype 3 increased pro- 202 (36%); ADNI: 31 (21%); P-values ranging between teins were enrichmed for acute inflammation, B-cell activa- 0.01  10–16 and 0.049]. The predominant cell types pro- tion, blood coagulation-related processes, lipid processing, ducing these proteins were oligodendrocytes, neurons and and lipoprotein clearance, which together suggest that this astrocytes. GO pathway analyses for proteins specifically subtype may be characterized as having blood–brain barrier increased in subtype 2 showed enrichment for innate im- dysfunction. Subtype 3 also showed enrichment for comple- mune response, extracellular matrix organization and oligo- ment activation, but for a different group of proteins than dendrocyte development, hence these individuals may be observed in subtype 2: C6, C8A, C8B and C9, which are characterized as having innate immune activation (54% of part of the terminal pathway of the complement system the processes enriched in ADNI overlapped with those (Veerhuis et al., 2011; Orsini et al., 2014) (Supplementary enriched in EMIF-AD MBD). Compared to controls, sub- Fig. 3). We searched the ENCODE and ChEA consensus type 3 individuals showed mostly decreased proteins [415 transcriptional factor database using the Enrichr server to 3782 | BRAIN 2020: 143; 3776–3792 B. M. Tijms et al. Figure 1 Cluster results. (A) Subject loadings on subtype scores (orange: subtype 1, hyperplasticity; blue: subtype 2, innate immune activa- tion; green: subtype 3, blood–brain barrier dysfunction) for EMIF-AD MBD (left) and ADNI (right). Each dot shows how an individual matches all three proteomic subtypes at the same time, e.g. the right-most green dot is a subject who shows very high loading on the subtype 3 axis, and very low loadings on subtypes 1 and 2 axes. (B) Heat map of subtype average Z-scores (according to the mean and SD of controls). Labels not shown, see Supplementary Table 5 for list of proteins. (C) Proportion of cell type production for protein levels higher (positive proportions) or lower than control subjects (negative proportions). (D) Selected subset of GO pathways that show subtype-specific enrichment with log(pFDR) positive values for proteins with higher levels than controls, and negative values for proteins with lower levels than controls (see Supplementary Table 7 for complete list of enriched pathways). Pathophysiological Alzheimer’s disease subtypes BRAIN 2020: 143; 3776–3792 | 3783 identify potential upstream drivers for proteins that were immune activation) were older and more often male. In both highest in each subtype and significantly different from con- cohorts, t-tau and p-tau CSF levels were highest and most trols. The strongest transcriptional factors associated with often abnormal in the subtype 1 (hyperplasticity; Fig. 3B increased proteins specifically in subtype 1 in both EMIF- and Supplementary Table 10), intermediate for subtype 2 AD MBD and ADNI were REST and SUZ12, which are (innate immune activation), and the lowest and most often transcriptional repressors that are important for neuronal normal in subtype 3 (blood–brain barrier dysfunction). development and synaptic plasticity (Rodenas-Ruano et al., Other neuronal injury markers such as neurogranin (both 2012) (Supplementary Table 8). The top transcription fac- cohorts), VILIP and SNAP25 (ADNI only) were consistently tors associated with subtype 2 did not show consistent pat- highest in subtype 1 (hyperplasticity), and lowest in subtype terns between EMIF-AD MBD and ADNI, and no 3 (blood–brain barrier dysfunction) (Fig. 3C and D). NEFL significant transcription factors were observed for subtype 3- levels were comparable across subtypes in EMIF-AD MBD, specific increases (tested in EMIF-AD MBD only). but were increased in subtype 2 (innate immune activation) Longitudinal proteomics was available for a subset in ADNI in ADNI, which remained after additional correction for age [n = 70 (29%), including 23 controls and 47 Alzheimer’s dis- and sex. Subtype 1 (hyperplasticity) further showed higher ease: 31 with subtype 1; nine with subtype 2; seven with levels of proteins associated with amyloid precursor protein subtype 3], including only proteins associated with subtypes (APP) processing (i.e. higher levels of amyloid-b1–40 and 1 and 2. Proteomic profile scores that summarized levels of amyloid-b1–38 in both cohorts, and higher levels of BACE1 proteins that were associated with either subtype 1 (52 pro- activity in ADNI). Subtype 3 (blood–brain barrier dysfunc- teins) or 2 (12 proteins) in the discovery dataset remained tion) showed the lowest concentrations for those markers. stable over time in all subtypes, as none of the slopes dif- Both subtypes 1 (hyperplasticity) and 2 (innate immune acti- fered from 0 (all subtypes P 4 0.10, Supplementary Table vation) showed higher levels of inflammation markers 9). This suggests that subtype definitions remained stable YKL40 and sTREM2 (ADNI only) than subtype 3 (blood– over time. brain barrier dysfunction). Since some of these markers can increase with disease severity, we repeated subtype compari- Genetic comparisons of subtypes sons stratified for disease stage (normal cognition, MCI and dementia). Results showed largely similar subtype profiles Subtypes showed similar proportions of APOE e4 carriers (all Pinteraction 4 0.05; Supplementary Figs 4 and 5). in both cohorts (Fig. 2A and Supplementary Table 10; all P’s 4 0.05). Relative to controls, all subtypes had an excess of Alzheimer’s disease genetic risk (Fig. 2B and Atrophy, vascular damage, cognitive Supplementary Table 11; P-values ranging between profiles and pathological 0.02  10–13 and 0.004). For SNP inclusion thresholds 0.1 comparisons to 0.5, subtype 2 individuals showed higher Alzheimer’s dis- ease polygenic risk score than subtype 1 and 3, but these Atrophy relative to controls was most pronounced in the associations lost significance after adjusting for age and sex. hippocampus, medial and lateral temporal cortex and the Because subtype 2 individuals were associated with innate precuneus for all subtypes (Fig. 4A, Supplementary Table 12 immune response, which has been previously associated and Supplementary Fig. 6). Compared to subtype 1 (hyper- with top Alzheimer’s disease risk SNPs, we compared sub- plasticity), individuals with subtypes 2 (innate immune acti- types on polygenic risk score for innate immune response vation) and 3 (blood–brain barrier dysfunction) showed and complement activation, and found for the majority of more atrophy in the posterior cingulate in both cohorts SNP inclusion thresholds the highest scores for subtype 2 (Fig. 4B). In ADNI, subtype 2 (innate immune activation) (Fig. 2C and Supplementary Table 11; P-values compared to showed more atrophy than subtype 1 (hyperplasticity) in the controls ranging between 0.02  10–7 and 0.045; P-values inferior temporal gyrus, insula, isthmus cingulate, rostral compared to the other subtypes ranging between 0.004 and middle frontal and temporal pole. Visual ratings for vascular 0.045). These effects remained largely unchanged after damage on MRI in EMIF-AD MBD showed that subtype 3 adjusting for age and sex. (blood–brain barrier dysfunction) more often had a lacunar infarct (n = 10, 22%) than subtype 2 (n = 1, 2%; P = 0.003) Other biological and clinical subtype and subtype 1 (n = 4, 7.5%; P = 0.04). No differences be- tween subtypes were observed in white matter intensity load characterization (Fazekas score of 3), or the presence of more than one We next compared subtypes on clinical characteristics and microbleed (Supplementary Table 10). In ADNI, white mat- established Alzheimer’s disease CSF markers. In EMIF-AD ter hyperintensity volumes were larger in subtype 3 (blood– MBD, subtypes had comparable proportions of disease brain barrier dysfunction; 1.2 ± 2.7 cm3) and subtype 2 (in- stages and sex, and comparable age (Fig. 3A). In ADNI, nate immune activation; 1.3 ± 1.4 cm3) compared to subtype individuals with subtype 1 (hyperplasticity) less often had 1 (hyperplasticity; 0.85 ± 3.0 cm3; subtype 1 versus subtype dementia (compared to subtype 2, P = 0.02; compared to 2, P = 0.0004; subtype 1 versus subtype 3, P = 0.01; sub- subtype 3, P = 0.02), and individuals with subtype 2 (innate type 2 versus subtype 3, P = 0.44). Subtypes showed largely 3784 | BRAIN 2020: 143; 3776–3792 B. M. Tijms et al. Figure 2 Genetic factor comparisons between subtypes. (A) Proportion of APOE e4 carriers according to subtype. (B) Effect sizes (95% CI) of Alzheimer’s disease polygenic risk scores for increasing SNP inclusion P-value thresholds, comparing Alzheimer’s disease subtypes to the control group (normal cognition, normal CSF amyloid and tau levels). (C) Comparisons of Alzheimer’s disease subtypes in Alzheimer’s disease polygenic risk specific for GO innate immune response (left) and GO complement activation (right), for increasing SNP inclusion P-value thresh- olds. (B and C) The pooled sample, and the cohorts separately. The dotted vertical lines in B and C indicate mean scores for the control group. See Supplementary Tables 10 and 11 for test statistics of all comparisons. similar scores on cognitive tests (Fig. 4C and Supplementary [HR = 2.1 (1.0, 4.4), P = 0.06] (Fig. 4G and Supplementary Table 13). Repeating analyses stratified on disease stage, Table 15). For a subset of 20 (10%) ADNI individuals with showed that individuals with subtype 3 in the dementia neuropathological information, we found similar pathologic- stage scored worse on the TMT (A) than the other two sub- al scores for amyloid and tau for subtypes, and they showed types (Pinteraction = 0.004; both subtype 1 versus subtype 3, similar frequencies of occurring co-pathologies, such as and subtype 2 versus subtype 3, P 5 0.001) (Supplementary Lewy body pathology, TDP-43 and hippocampal sclerosis Table 13 and Supplementary Fig. 7). Worsening over time (Supplementary Table 16). on Clinical Dementia Rating Scale Sum of Boxes (CDRsob) was steeper for subtype 2 compared to subtype 1 in MCI (P = 0.01; Fig. 4E, F and Supplementary Table 14), and for subtype 3 compared to subtype 1 in dementia (P = 0.02). Discussion Compared to subtype 1, individuals without dementia and Understanding biological heterogeneity in patients with subtype 2 showed increased risk of progression to dementia, Alzheimer’s disease is critical for treatment development. We also after correcting for age, sex, level of education and tau proteomically defined three Alzheimer’s disease pathophysio- levels [hazard ratio (HR) (95% CI) subtype 2 versus subtype logical subtypes that were associated with distinct biological 1 = 2.5 (1.2, 5.1), P = 0.01], and subtype 3 at trend level processes, i.e. hyperplasticity, innate immune activation and Pathophysiological Alzheimer’s disease subtypes BRAIN 2020: 143; 3776–3792 | 3785 Figure 3 Clinical stage, age, sex and other CSF markers comparisons between subtypes. (A) Proportions of disease stage (NC = normal cognition), and females according to Alzheimer’s disease subtypes and cohort. (B) Distributions of age, t-tau and p-tau levels according to Alzheimer’s disease subtypes and cohort. (C) Distributions of CSF markers not included in clustering according to subtype and cohort, including additional CSF biomarkers that were available in ADNI only for a subset of individuals. See Supplementary Table 10 for test statistics of all comparisons. 3786 | BRAIN 2020: 143; 3776–3792 B. M. Tijms et al. Figure 4 Cortical thickness and cognition comparisons between subtypes. (A) All brain areas where differences compared to controls were observed. (B) Brain areas with a significant main effect for subtype; all beta values reflect volumetric differences of subtypes compared to controls. (C) Comparisons of cognitive profiles between subtypes. (D) Changes over time on MMSE (left), and CDRsob (right) in ADNI only. (E) Cumulative progression to dementia curves for subtypes, in ADNI only. All cortical thickness and neuropsychological test values are standardized according to the mean and SD values of the control subjects. See Supplementary Tables 10 and 12–15 for test statistics of all comparisons. Pathophysiological Alzheimer’s disease subtypes BRAIN 2020: 143; 3776–3792 | 3787 blood–brain barrier dysfunction, and these subtypes were ro- REST and SUZ12, which are transcription suppressors that bustly observed in two large independent cohorts. These bio- are known to interact with each other (Tsai et al., 2010) and logical subtypes of Alzheimer’s disease showed pronounced play an important role in regulating neuronal development differences in levels of proteins associated with processes and synaptic plasticity (Ballas and Mandel, 2005; Rodenas- known to be deregulated in Alzheimer’s disease, including Ruano et al., 2012; Yang et al., 2012). These factors have APP processing, neuronal injury, and inflammation. All sub- previously been identified as key regulators in a sporadic type-specific alterations in CSF protein levels could both be Alzheimer’s disease iPSC model (Meyer et al., 2019), in increased or decreased for subsets of proteins, indicating which neurons also showed increase excitability, and that differences cannot be explained by trivial aspecific increased tau phosphorylation and amyloid-b1–40 concentra- changes in CSF composition. All subtypes had an excess gen- tions, which is in line with our observations and interpret- etic risk for Alzheimer’s disease, and pathological measures ation of subtype 1. Finally, proteins increased in subtype 1 did not show a difference in the presence of comorbidities, also showed enrichment for glucose metabolism, which has providing further support that these differences reflect het- been reported in two recent proteomic studies in tissue and erogeneity within Alzheimer’s disease. A particularly novel in CSF that also observed clusters of proteins associated finding is the observation of the blood–brain barrier dys- with glucose metabolism to be involved in Alzheimer’s dis- function subtype that showed mostly abnormally low con- ease, as well as clusters of synaptic proteins (Higginbotham centrations of proteins associated with APP processing, as et al., 2019; Johnson et al., 2020). In those studies, and our well as t-tau and p-tau levels. Together, these results demon- results, glucose metabolism pathways mostly included pro- strate the potential for CSF proteomics to identify which bio- teins that were specific for neurons and/or astrocytes. logical processes are disrupted in individual patients with Neurons can increase neuronal glycolysis during neuronal Alzheimer’s disease, and suggest that individuals might re- stimulation (Dı́az-Garcı́a et al., 2017). Although neurons use quire specific treatments depending on their subtype. most of the brain’s energy, glucose metabolism is predomin- Previous studies that clustered targeted proteins amyloid, antly driven by astrocytes, which themselves consume only t-tau, p-tau and/or ubiquitin CSF levels suggested three to up to 15% of energy in the brain, and can regulate glucose five Alzheimer’s disease subtypes that were characterized by metabolism upon neuronal activation (Magistretti and having low, intermediate and high tau values (Iqbal et al., Allaman, 2015). These findings indicate that non-neuronal 2005; van der Vlies et al., 2009; Wallin et al., 2010), and cell types may be dysregulated and/or have detrimental our proteomically defined subtypes show a similar distinc- effects on neurons in subtype 1. tion in tau levels. We further show, with our large-scale The second subtype had a proteomic profile that also indi- proteomic analyses, which biological processes may underlie cated upregulation of plasticity-related processes like subtype interindividual differences in tau levels. Individuals with 1, but less pronounced. Subtype 2 further showed higher lev- Alzheimer’s disease with the hyperplasticity subtype showed els of proteins that—in both cohorts—pointed to involvement high levels for the majority of proteins, and pathways of the innate immune system, extracellular matrix organiza- enriched in both cohorts showed involvement of regulation tion and oligodendrocyte development. Oligodendrocytes are of the MAPK/ERK cascade, glucose metabolism, synaptic important for axonal myelination. In ADNI, subtype 2 structure and function, and axonal development, all proc- showed increased white matter hyperintensity volume, and esses important for synaptic plasticity. This hyperplasticity was increased in both EMIF-AD MBD and ADNI NEFL, sug- subtype also showed higher levels of markers presumed to gesting axonal damage in this subtype. Furthermore, individu- reflect neuronal injury, i.e. t-tau, p-tau, and neurogranin in als with subtype 2 showed high levels of complement proteins both cohorts (as well as VILIP, and SNAP25, which was C1q B chain and C4a in both cohorts, as well as C1q A only measured in ADNI), which could reflect more severe chain, C1q C chain, C1s, and C1r in EMIF-AD MBD (not neuronal damage (Fagan and Perrin, 2012; Brinkmalm measured in ADNI), which are early components of the clas- et al., 2014). However, this is unlikely because these proteins sical complement pathway (Veerhuis et al., 2011; Orsini were already increased in individuals with Alzheimer’s dis- et al., 2014). Alzheimer’s disease PGRSs restricted to genes ease and normal cognition, when atrophy was less severe involved in innate immune response and complement activa- compared to the other subtypes. An alternative explanation tion were mostly higher in subtype 2 in both cohorts, suggest- could be increased synaptic activity as this can lead to ing that these CSF proteomic alterations reflect genetic effects. increased tau (Pooler et al., 2013; Yamada et al., 2014) and Higher concentrations of C1q and C4 in Alzheimer’s disease amyloid secretion (Cirrito et al., 2005; Bero et al., 2011). brains have been reported in pathological studies (Veerhuis Hyperactive neurons have been reported in Alzheimer’s dis- et al., 2011; Dejanovic et al., 2018), and therefore higher con- ease with concurrent increased tau and amyloid levels centrations of C4a may indicate complement activation in this (Roberson et al., 2011; Palop and Mucke, 2016). Aberrant subtype. Amyloid-b fibrils are known to activate the comple- increases in neuronal activation can be caused by amyloid ment pathway by binding to the C1q complex (Rogers et al., oligomers, which disrupt the balance of excitation and inhib- 1992; Webster et al., 2002). Complement activation may also ition of neuronal circuits (Palop and Mucke, 2016). play a role in neuronal injury in Alzheimer’s disease, because Furthermore, subtype 1-specific proteomic increases in both complement proteins can accumulate at synapses and tag EMIF and ADNI converged on the transcription factors these for phagocytosis by activated microglia (Pooler et al., 3788 | BRAIN 2020: 143; 3776–3792 B. M. Tijms et al. 2013; Orsini et al., 2014; Yamada et al., 2014). Knocking specifically increased in subtype 3 (in EMIF-AD MBD out C1q in APP transgenic mice or blocking C1q in Tau- only) was enriched for lipid processing, clearance and P301S transgenic mice attenuates both complement activation regulation, and these included apoC1. ApoC1 is produced and neuronal injury (Pooler et al., 2013; Orsini et al., 2014; by astrocytes (Petit-Turcotte et al., 2001; Abildayeva Yamada et al., 2014), possibly by preventing inappropriate et al., 2008), can inhibit receptor-mediated clearance of microglia activation. Together, the biological processes specif- lipoproteins containing APOE (Sehayek and Eisenberg, ic for subtype 2 seem to be associated with activated micro- 1991; Shachter, 2001), and has been observed in amyloid glia, which may contribute to neuronal dysfunction. plaques (Abildayeva et al., 2008). This suggests that vas- Alternatively, these processes may be associated with activa- cular factors might play a role in amyloid pathogenesis, tion or dysregulation of astrocytes, because microglia secret- possibly contributing to reduced clearance of aggregated ing C1q can induce so-called ‘A1 reactive’ astrocytes that lose amyloid. Alternatively, amyloid might aggregate in the the ability to facilitate plasticity processes that promote cell vasculature, which could lead to blood–brain barrier survival and accelerate death of neurons and oligodendrocytes dysfunction. (Liddelow et al., 2017). The proteomic subtypes we discovered could have implica- In both cohorts, subtype 3 had low and more often nor- tions for treatment: subtype 1 showed the highest levels of mal t-tau and p-tau CSF levels compared to the other sub- BACE1 activity (measured in ADNI only) and products of types, together with abnormally low levels for the amyloid metabolism (amyloid-b1–40 and amyloid-b1–38; majority of other proteins. The generally low levels of t- measured in both cohorts) and so it can be hypothesized and p-tau raise the question as to whether these individu- that particularly this subtype will benefit from treatments als have Alzheimer’s disease (Jack et al., 2018). Our that target APP processing, such as BACE1 inhibitors, results provide strong support that in fact these individu- whereas this type of treatment may be harmful for individu- als do have Alzheimer’s disease, because: (i) these individ- als with subtype 3 that showed decreased levels of BACE1 uals had abnormal amyloid levels; (ii) the most severe activation. Individuals with subtype 2 may potentially bene- atrophy included typical Alzheimer’s disease regions, such fit from therapeutic strategies that target microglia and as the medial temporal lobe; (iii) non-demented individu- astrocyte activation. Subtype 3 may benefit from therapies als with subtype 3 showed increased risk for clinical pro- that protect the vasculature. Future research should further gression; and (iv) they have an excess of genetic risk for study treatment effects on CSF proteomic profiles, and Alzheimer’s disease. The percentage of individuals classi- whether effects are subtype-dependent. fied as having subtype 3 is in line with previous studies A potential limitation of this study is that proteins specific- reporting pathologically confirmed Alzheimer’s disease, ally increased in subtype 3 were mostly observed in the which show that up to 30% of individuals can have nor- EMIF-AD MBD cohort, because that study used an untar- mal CSF tau levels (Shaw et al., 2009). Thus, CSF tau lev- geted approach, whereas those proteins were not measured els may reflect other processes in addition to in ADNI, because that study selected a limited number of neurofibrillary tau tangles, and conversely, normal levels proteins with brain enriched expression patterns. However, do not exclude underlying tau pathology. Given the rela- subtype 3 in ADNI showed a hypoplasticity response, simi- tionship of tau levels and neuronal activity discussed lar to that of subtype 3 in EMIF-AD MBD, suggesting that above, low tau levels in this subtype may reflect hypoplas- they share common pathophysiological processes. ticity. Alternatively, low levels of tau suggest less neuronal Furthermore, the random forest analysis showed that in injury. However, this explanation seems implausible, be- principle, it is possible to use these proteins as robust sub- cause subtype 3 had—in both cohorts—widespread atro- type markers, independent of cohort or measuring tech- phy and significantly higher levels of the axonal damage nique. Still, further analyses focussing on single biomarkers marker NEFL. Furthermore, in EMIF-AD MBD only (be- are necessary since in ADNI, proteins were preselected, while cause that cohort used an untargeted proteomics tech- in EMIF, an untargeted technique was used, and so not all nique), a large group of proteins that were increased in top markers in EMIF for specific subtypes were available in subtype 3 have previously been reported to correlate with ADNI. Also, new studies are necessary for further validation the CSF/plasma albumin ratio, which is a marker for of these subtypes, using independent sets of proteins for sub- blood–brain barrier integrity (Dayon et al., 2019), and typing and for characterization to avoid potential inflation this suggests that subtype 3 may have blood–brain barrier of classification accuracy due to circularity. Furthermore, al- dysfunction (Sagare et al., 2012; Yamazaki and Kanekiyo, though many of the proteins specifically increased in subtype 2017; Sweeney et al., 2018). Blood–brain barrier dysfunc- 3 were previously reported to be correlated to blood–brain tion disrupts glucose metabolism, which can impair neur- barrier function (Dayon et al., 2019), future analyses should onal activity and plasticity processes (Yamazaki and further verify this by measuring both CSF and plasma albu- Kanekiyo, 2017; Sweeney et al., 2018). This would ex- min. Furthermore, the EMIF-AD MBD and ADNI cohorts plain why proteins involved in synaptic structure and are a clinical multicentre and a research multicentre study, function were decreased in subtype 3, consistently respectively, and differed in their baseline characteristics, observed in both cohorts, and suggest that these individu- most notably ADNI participants being on average 10 years als have hypoplasticity. Another subset of proteins older. However, it is unlikely cohort-specific and age effects Pathophysiological Alzheimer’s disease subtypes BRAIN 2020: 143; 3776–3792 | 3789 explain the subtypes, as the subtypes showed highly similar and the Vlaamse Impulsfinanciering voor Netwerken voor proteomic profiles across the cohorts. Furthermore, MRI Dementie-onderzoek (IWT #135043). scans in EMIF-AD MBD were acquired in clinical routine, Data were used for this project of which collection and in contrast to the harmonized scanning protocol in ADNI. sharing was funded by the Alzheimer’s Disease This may have made it more difficult to detect differences in Neuroimaging Initiative (ADNI) (National Institutes of atrophy patterns amongst subtypes in combination with Health Grant U01 AG024904) and DOD ADNI relatively small sample sizes (Zhang et al., 2016; Scheltens (Department of Defense award number W81XWH-12-2- et al., 2017; Ten Kate et al., 2018). Another point of consid- 0012). ADNI is funded by the National Institute on Aging, eration is that co-pathology often occurs with older age, the National Institute of Biomedical Imaging and which may influence proteomic subtype definitions (Beach Bioengineering, and through generous contributions from et al., 2012). Our analyses in a subset of the ADNI cohort the following: AbbVie, Alzheimer’s Association; Alzheimer’s showed that subtypes did not differ in the occurrence of co- Drug Discovery Foundation; Araclon Biotech; BioClinica, pathology, suggesting that it is unlikely that this has driven Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; the subtypes. Still, those analyses need further replication in Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and larger samples with combined tissue and CSF proteomic Company; EuroImmun; F. Hoffmann-La Roche Ltd and its data. Our subtypes were defined by their proteomic profiles, affiliated company Genentech, Inc.; Fujirebio; GE and at this point single markers from our analyses should Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy not be used in practice until thoroughly validated, which we Research & Development, LLC.; Johnson & Johnson aim to pursue in future studies. A strength of our study is Pharmaceutical Research & Development LLC.; Lumosity; that we were able to replicate the biological subtypes that Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, we detected with CSF proteomics in two independent LLC.; NeuroRx Research; Neurotrack Technologies; cohorts, and even though different methods were used to Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal measure proteins, we observed similar processes to be Imaging; Servier; Takeda Pharmaceutical Company; and involved in Alzheimer’s disease, supporting the robustness of Transition Therapeutics. The Canadian Institutes of Health our findings. Research is providing funds to support ADNI clinical sites in In conclusion, we have identified a hyperplasticity, innate Canada. Private sector contributions are facilitated by the immune activation and a blood–brain barrier dysfunction Foundation for the National Institutes of Health (www.fnih. subtype in Alzheimer’s disease using CSF proteomics. The org). The grantee organization is the Northern California most important implication of our results is that currently Institute for Research and Education, and the study is coor- existing—and even failed—treatments may be beneficial for dinated by the Alzheimer’s Therapeutic Research Institute at specific subtypes, and that CSF proteomics may serve as a the University of Southern California. ADNI data are disse- stratification tool to investigate this further. minated by the Laboratory for Neuro Imaging at the University of Southern California. Funding This work has been supported by ZonMW Memorabel Competing interests grant programme #73305056 (B.M.T.) and #733050824 B.M.T., P.J.V., J.G., H.Z. and K.B. are the inventors of the (B.M.T. and P.J.V.), the Swedish Research Council (#2018- CSF proteomic profile definition of Alzheimer’s disease sub- 02532, H.Z.), the European Research Council (#681712, types and have a patent pending (#19165795.6; Applicant: H.Z.) and Swedish State Support for Clinical Research Stichting VUmc). H.Z. has served at scientific advisory (#ALFGBG-720931, H.Z.), the Alzheimerfonden (#AF- boards for Roche Diagnostics, Wave, Samumed and CogRx, 930934, J.G.) and Dtiftelsen Gamla tjänarirror (J.G.), and has given lectures in symposia sponsored by Biogen and the Innovative Medicines Initiative Joint Undertaking under Alzecure, and is a co-founder of Brain Biomarker Solutions EMIF grant agreement #115372 (P.J.V., H.Z.). Statistical in Gothenburg AB, a GU Ventures-based platform company analyses were performed at the VUmc Alzheimer Center that at the University of Gothenburg (all unrelated to the submit- is part of the neurodegeneration research program of the ted work). S.L. is currently an employee at Janssen R&D Neuroscience Campus Amsterdam. EMIF-AD MBD prote- and has in the past five years provided consultancy to Eisai, omic analyses were performed at the Department of SomaLogic, Merck and Optum Labs. F.B. is a consultant for Psychiatry and Neurochemistry, the Sahlgrenska Academy Biogen, Bayer, Merck, Roche, Novartis, Lundbeck, and at the University of Gothenburg, Sweden. The VUmc IXICO; has received sponsoring from European Alzheimer Center is supported by Stichting Alzheimer Commission–Horizon 2020, National Institute for Health Nederland and Stichting VUmc fonds. H.Z. is a Wallenberg Research–University College London Hospitals Biomedical Academy Fellow. F.B. is supported by the NIHR biomedical Research Centre, Biogen, TEVA and Novartis. The other research centre at UCLH. The Leuven cohort was funded by authors declare no competing interests with the content of Stichting Alzheimer Onderzoek (#11020, #15005, #13007) this article. 3790 | BRAIN 2020: 143; 3776–3792 B. M. Tijms et al. disease: revising the NINCDS–ADRDA criteria. Lancet Neurol Supplementary material 2007; 6: 734–46. Dubois B, Feldman HH, Jacova C, Hampel H, Molinuevo JL, Supplementary material is available at Brain online. Blennow K, et al. Advancing research diagnostic criteria for Alzheimer’s disease: the IWG-2 criteria. Lancet Neurol 2014; 13: 614–29. References Duits FH, Brinkmalm G, Teunissen CE, Brinkmalm A, Scheltens P, van der Flier WM, et al. Synaptic proteins in CSF as potential novel Abildayeva K, Berbée JFP, Blokland A, Jansen PJ, Hoek FJ, Meijer O, biomarkers for prognosis in prodromal Alzheimer’s disease. et al. Human apolipoprotein C-I expression in mice impairs learning Alzheimers Res Ther 2018; 10: 387. and memory functions. J Lipid Res 2008; 49: 856–69. Euesden J, Lewis CM, O’Reilly PF. PRSice: polygenic Risk Score soft- Albert MS, DeKosky ST, Dickson D, Dubois B, Feldman HH, Fox ware. Bioinformatics 2014; 31: 1466–8. NC, et al. The diagnosis of mild cognitive impairment due to Fagan AM, Perrin RJ. Upcoming candidate cerebrospinal fluid bio- Alzheimer’s disease: recommendations from the National Institute markers of Alzheimer’s disease. Biomarkers Med 2012; 6: 455–76. on Aging-Alzheimer’s Association workgroups on diagnostic guide- Gaujoux R, Seoighe C. A flexible R package for nonnegative matrix lines for Alzheimer’s disease. Alzheimers Dement 2011; 7: 270–9. factorization. BMC Bioinformatics 2010; 11: 367. Ballas N, Mandel G. The many faces of REST oversee epigenetic pro- Higginbotham L, Ping L, Dammer E, Duong D, Zhou M, Gearing M, gramming of neuronal genes. Curr Opin Neurobiol 2005; 15: et al. Integrated proteomics reveals brain-based cerebrospinal fluid 500–6. biomarkers in asymptomatic and symptomatic Alzheimer’s Disease. Batth TS, Francavilla C, Olsen JV. Off-line high-pH reversed-phase bioRxiv 2019; 806752. doi:10.1101/806752. fractionation for in-depth phosphoproteomics. J Proteome Res Hondius DC, van Nierop P, Li KW, Hoozemans JJM, van der Schors 2014; 13: 6176–86. RC, van Haastert ES, et al. Profiling the human hippocampal prote- Beach TG, Monsell SE, Phillips LE, Kukull W. Accuracy of the clinical ome at all pathologic stages of Alzheimer’s disease. Alzheimers diagnosis of Alzheimer Disease at National Institute on Aging Dement 2016; 12: 654–68. Alzheimer Disease Centers, 2005–2010. J Neuropathol Exp Neurol Hong S, Prokopenko D, Dobricic V, Kilpert F, Bos I, Vos SJB, et al. 2012; 71: 266–73. Genome-wide association study of Alzheimer’s disease CSF bio- Bero AW, Yan P, Roh JH, Cirrito JR, Stewart FR, Raichle ME, et al. markers in the EMIF-AD Multimodal Biomarker Discovery dataset. Neuronal activity regulates the regional vulnerability to amyloid-b bioRxiv 2019; 1: 412–36. deposition. Nat Neurosci 2011; 14: 750–6. Iqbal K, Flory M, Khatoon S, Soininen H, Pirttila T, Lehtovirta M, et Bertens D, Tijms BM, Scheltens P, Teunissen CE, Visser PJ. Unbiased al. Subgroups of Alzheimer’s disease based on cerebrospinal fluid estimates of cerebrospinal fluid b-amyloid 1–42 cutoffs in a large molecular markers. Ann Neurol 2005; 58: 748–57. memory clinic population. Alzheimers Res Ther 2017; 9: 614. Jack CR Jr, Albert MS, Knopman DS, Mckhann GM, Sperling RA, Blennow K, Wallin A. Clinical heterogeneity of probable Alzheimer’s Carrillo MC, et al. Introduction to the recommendations from the disease. J GeriatrPsychiatry Neurol 1992; 5: 106–13. National Institute on Aging-Alzheimer‘s Association workgroups on Bos I, Vos S, Vandenberghe R, Scheltens P, Engelborghs S, Frisoni G, diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement et al. The EMIF-AD Multimodal Biomarker Discovery study: design, 2011; 7: 257–62. methods and cohort characteristics. Alzheimers Res Therapy 2018; Jack CR Jr, Bennett DA, Blennow K, Carrillo MC, Dunn B, 10: 207. Haeberlein SB, et al. NIA-AA Research Framework: toward a bio- Brinkmalm A, Brinkmalm G, Honer WG, Frölich L, Hausner L, logical definition of Alzheimer’s disease. Alzheimers Dement 2018; Minthon L, et al. SNAP-25 is a promising novel cerebrospinal fluid 14: 535–62. biomarker for synapse degeneration in Alzheimer’s disease. Mol Jansen IE, Savage JE, Watanabe K, Bryois J, Williams DM, Steinberg Neurodegener 2014; 9: 53. S, et al. Genome-wide meta-analysis identifies new loci and function- Cirrito JR, Yamada KA, Finn MB, Sloviter RS, Bales KR, May PC, et al pathways influencing Alzheimer’s disease risk. Nat Genet 2020; al. Synaptic activity regulates interstitial fluid amyloid-b levels 51: 404–13. in vivo. Neuron 2005; 48: 913–22. Johnson ECB, Dammer EB, Duong DM, et al. Large-scale proteomic Chen EY, Tan CM, Kou Y, et al. Enrichr: interactive and collaborative analysis of Alzheimer’s disease brain and cerebrospinal fluid reveals HTML5 gene list enrichment analysis tool. BMC Bioinformatics early changes in energy metabolism associated with microglia and 2013; 14: 128–14. astrocyte activation. Nat Med 2020; 26: 769–80. Dayon L, Cominetti O, Wojcik J, Galindo AN, Oikonomidi A, Henry Kuleshov MV, Jones MR, Rouillard AD, et al. Enrichr: a comprehen- H, et al. Proteomes of paired human cerebrospinal fluid and plasma: sive gene set enrichment analysis web server 2016 update. Nucleic relation to blood–brain barrier permeability in older adults. J Acids Res 2016; 44: W90–W97. Proteome Res 2019; 18: 1162–74. Kunkle BW, Grenier-Boley B, Sims R, Bis JC, Damotte V, Naj AC, et De Leon MJ, Pirraglia E, Osorio RS, Glodzik L, Saint-Louis L, Kim al. Genetic meta-analysis of diagnosed Alzheimer’s disease identifies HJ, et al. The nonlinear relationship between cerebrospinal fluid new risk loci and implicates Ab, tau, immunity and lipid processing. Ab42 and tau in preclinical Alzheimer’s disease. PLoS One 2018; Nat Genet 2019; 51: 414–30. 13: e0191240. Lam B, Masellis M, Freedman M, Stuss DT, Black SE. Clinical, imag- De Meyer G. Diagnosis-independent Alzheimer disease Biomarker signa- ing, and pathological heterogeneity of the Alzheimer’s disease syn- ture in cognitively normal elderly people. Arch Neurol 2010; 67: 949. drome. Alzheimers Res Ther 2013; 5:1. Dejanovic B, Huntley MA, De Mazière A, Meilandt WJ, Wu T, Lambert J-C, Ibrahim-Verbaas CA, Harold D, Naj AC, Sims R, Srinivasan K, et al. Changes in the synaptic proteome in tauopathy Bellenguez C, et al. Meta-analysis of 74,046 individuals identifies 11 and rescue of tau-induced synapse loss by C1q antibodies. Neuron new susceptibility loci for Alzheimer’s disease. Nat Genet 2013; 45: 2018; 100: 1322–36.e7. 1452–8. doi: 10.1038/ng.2802. Dı́az-Garcı́a CM, Mongeon R, Lahmann C, Koveal D, Zucker H, Lee DD, Seung HS. Learning the parts of objects by non-negative ma- Yellen G. Neuronal stimulation triggers neuronal glycolysis and not trix factorization. Nature 1999; 401: 788–91. lactate uptake. Cell Metab 2017; 26: 361–74. Liddelow SA, Guttenplan KA, Clarke LE, Bennett FC, Bohlen CJ, Dubois B, Feldman HH, Jacova C, DeKosky ST, Barberger-Gateau P, Schirmer L, et al. Neurotoxic reactive astrocytes are induced by acti- Cummings J, et al. Research criteria for the diagnosis of Alzheimer’s vated microglia. Nature 2017; 541: 481–7. Pathophysiological Alzheimer’s disease subtypes BRAIN 2020: 143; 3776–3792 | 3791 Maarouf C, Andacht T, Kokjohn T, Castano E, Sue L, Beach T, et al. Rogers J, Cooper NR, Webster S, Schultz J, McGeer PL, Styren SD, et Proteomic analysis of Alzheimers disease cerebrospinal fluid from al. Complement activation by beta-amyloid in Alzheimer disease. neuropathologically diagnosed subjects. Curr Alzheimer Res 2009; Proc Natl Acad Sci USA 1992; 89: 10016–10020. 6: 399–406. Sagare AP, Bell RD, Zlokovic BV. Neurovascular dysfunction and Magdalinou NK, Noyce AJ, Pinto R, Lindstrom E, Holmén-Larsson J, faulty amyloid-peptide clearance in Alzheimer disease. Cold Spring Holtta M, et al. Identification of candidate cerebrospinal fluid bio- Harb Perspect Med 2012; 2: a011452. markers in parkinsonism using quantitative proteomics. Saykin AJ, Shen L, Foroud TM, Potkin SG, Swaminathan S, Kim S, et Parkinsonism Relat Disord 2017; 37: 65–71. al. Alzheimer’s disease Neuroimaging Initiative biomarkers as quan- Magistretti PJ, Allaman I. A cellular perspective on brain energy me- titative phenotypes: genetics core aims, progress, and plans. tabolism and functional imaging. Neuron 2015; 86: 883–901. Alzheimer’s Dement 2010; 6: 265–273. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan Scheltens NME, Tijms BM, Koene T, Barkhof F, Teunissen CE, EM. Clinical diagnosis of Alzheimer‘s disease: report of the Wolfsgruber S, et al. Cognitive subtypes of probable Alzheimer’s dis- NINCDS-ADRDA Work Group under the auspices of Department ease robustly identified in four cohorts. Alzheimers Dement 2017; of Health and Human Services Task Force on Alzheimer’s Disease. 13: 1226–1236. Neurology 1984; 34: 939. Sehayek E, Eisenberg S. Mechanisms of inhibition by apolipoprotein C McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR, Jr., of apolipoprotein E-dependent cellular metabolism of human trigly- Kawas CH, et al. The diagnosis of dementia due to Alzheimer’s dis- ceride-rich lipoproteins through the low density lipoprotein receptor ease: recommendations from the National Institute on Aging- pathway. J Biol Chem 1991; 266: 18259–18267. Alzheimer’s Association workgroups on diagnostic guidelines for Shachter NS. Apolipoproteins C-I and C-III as important modulators Alzheimer’s disease. Alzheimers Dement 2011; 7: 263–269. of lipoprotein metabolism. Curr Opin Lipidol 2001; 12: 297–304. Meyer K, Feldman HM, Lu T, et al. REST and neural gene network Shaw LM, Vanderstichele H, Knapik-Czajka M, Clark CM, Aisen PS, dysregulation in iPSC models of Alzheimer’s disease. Cell Rep 2019; Petersen RC, et al. Cerebrospinal fluid biomarker signature in 26: 1112–1127. Alzheimer’s disease neuroimaging initiative subjects. Ann Neurol Meyer P-F, Savard M, Poirier J, Labonte A, Rosa-Neto P, Weitz TM, 2009; 65: 403–413. et al. Bi-directional association of cerebrospinal fluid immune Smits LL, Pijnenburg YAL, van der Vlies AE, Koedam ELGE, Bouwman markers with stage of Alzheimer’s disease pathogenesis. J FH, Reuling IEW, et al. Early onset APOE E4-negative Alzheimer’s Alzheimers Dis 2018; 63: 577–590. disease patients show faster cognitive decline on non-memory Mi H, Muruganujan A, Casagrande JT, Thomas PD. Large-scale gene domains. Eur Neuropsychopharmacol 2015; 25: 1010–1017. function analysis with the PANTHER classification system. Nat Spellman DS, Wildsmith KR, Honigberg LA, Tuefferd M, Baker D, Protoc 2013; 8: 1551–1566. Raghavan N, et al. Development and evaluation of a multiplexed Möller C, Vrenken H, Jiskoot L, Versteeg A, Barkhof F, Scheltens P, et mass spectrometry based assay for measuring candidate peptide bio- al. Different patterns of gray matter atrophy in early- and late-onset markers in Alzheimer’s Disease Neuroimaging Initiative (ADNI) Alzheimer’s disease. Neurobiol Aging 2013; 34: 2014–2022. CSF. Proteomics Clin Appl 2015; 9: 715–731. Orsini F, De Blasio D, Zangari R, Zanier ER, De Simoni M-G. Sperling RA, Aisen PS, Beckett LA, Bennett DA, Craft S, Fagan AM, et Versatility of the complement system in neuroinflammation, neurode- al. Toward defining the preclinical stages of Alzheimer’s disease: rec- generation and brain homeostasis. Front Cell Neurosci 2014; 8: 380. ommendations from the National Institute on Aging-Alzheimer‘s Ossenkoppele R, Cohn-Sheehy BI, La Joie R, Vogel JW, Möller C, Association workgroups on diagnostic guidelines for Alzheimer’s Lehmann M, et al. Atrophy patterns in early clinical stages across disease. Alzheimers Dement 2011; 7: 280–292. distinct phenotypes of Alzheimer’s disease. Hum Brain Mapp 2015; Sweeney MD, Sagare AP, Zlokovic BV. Blood–brain barrier break- 36: 4421–4437. down in Alzheimer disease and other neurodegenerative disorders. Palop JJ, Mucke L. Network abnormalities and interneuron dysfunc- Nat Rev Neurol 2018; 14: 133–150. tion in Alzheimer disease. Nat Rev Neurosci 2016; 17: 777–792. Ten Kate M, Dicks E, Visser PJ, Van Der Flier WM, Teunissen CE, Pedrero-Prieto CM, Garcı́a-Carpintero S, Fronti~ nán-Rubio J, et al. A Barkhof F, et al. Atrophy subtypes in prodromal Alzheimer’s disease comprehensive systematic review of CSF proteins and peptides that are associated with cognitive decline. Brain 2018; 141: 3443–56. define Alzheimer’s disease. Clin Proteom 2020; 17: 21. doi: 10.1093/brain/awy264. Perez-Riverol Y, Csordas A, Bai J, et al. The PRIDE database and Tijms BM, Willemse EAJ, Zwan MD, Mulder SD, Visser PJ, van related tools and resources in 2019: improving support for quantifi- Berckel BNM, et al. Unbiased approach to counteract upward drift cation data. Nucleic Acids Res 2018; 47: D442–D450. in cerebrospinal fluid amyloid-b 1–42 Analysis Results. Clin Chem Petersen RC, Smith GE, Waring SC, Ivnik RJ, Tangalos EG, Kokmen 2018; 64: 576–585. E. Mild cognitive impairment: clinical characterization and outcome. Toledo JB, Xie SX, Trojanowski JQ, Shaw LM. Longitudinal change Arch Neurol 1999; 56: 303. in CSF Tau and Ab biomarkers for up to 48 months in ADNI. Acta Petit-Turcotte C, Stohl SM, Beffert U, Cohn JS, Aumont N, Tremblay Neuropathol 2013; 126: 659–670. M, et al. Apolipoprotein C-I expression in the brain in Alzheimer’s Tsai M-C, Manor O, Wan Y, et al. Long noncoding RNA as modular disease. Neurobiol Disease 2001; 8: 953–963. scaffold of histone modification complexes. Science 2010; 329: Pooler AM, Phillips EC, Lau DHW, Noble W, Hanger DP. 689–693. Physiological release of endogenous tau is stimulated by neuronal van der Vlies AE, Verwey NA, Bouwman FH, Blankenstein MA, Klein activity. EMBO Rep 2013; 14: 389–394. M, Scheltens P, et al. CSF biomarkers in relationship to cognitive Ridge PG, Hoyt KB, Boehme K, Mukherjee S, Haines JL, Mayeux R, profiles in Alzheimer disease. Neurology 2009; 72: 1056–1061. et al. Assessment of the genetic variance of late-onset Alzheimer’s Veerhuis R, Nielsen HM, Tenner AJ. Complement in the brain. Mol disease. Neurobiol Aging 2016; 41: 200.e13–200.e20. Immunol 2011; 48: 1592–1603. Roberson ED, Halabisky B, Yoo JW, Yao J, Chin J, Yan F, et al. Wallin AK, Blennow K, Zetterberg H, Londos E, Minthon L, Hansson Amyloid-/Fyn-induced synaptic, network, and cognitive impairments O. CSF biomarkers predict a more malignant outcome in Alzheimer depend on tau levels in multiple mouse models of Alzheimer’s dis- disease. Neurology 2010; 74: 1531–1537. ease. J Neurosci 2011; 31: 700–711. Webster S, Bradt B, Rogers J, Cooper N. Aggregation state-dependent Rodenas-Ruano A, Chávez AE, Cossio MJ, Castillo PE, Zukin RS. activation of the classical complement pathway by the amyloid b REST-dependent epigenetic remodeling promotes the developmental peptide. J Neurochem 2002; 69: 388–398. switch in synaptic NMDA receptors. Nat Neurosci 2012; 15: Wesenhagen KEJ, Teunissen CE, Visser PJ, Tijms BM. 1382–1390. Cerebrospinal fluid proteomics and biological heterogeneity in 3792 | BRAIN 2020: 143; 3776–3792 B. M. Tijms et al. Alzheimer’s disease: a literature review. Crit Rev Clin Lab Sci Yamazaki Y, Kanekiyo K. Blood-brain barrier dysfunction and the patho- 2020; 57: 86–98. genesis of Alzheimer’s Disease. Int J Mol Sci 2017; 18: 1965–19. Whitwell J, Dickson D, Murray M, Petersen R, Jack C, Josephs K. Yang YJ, Baltus AE, Mathew RS, et al. Microcephaly gene links tri- MRI in pathologically-defined hippocampal sparing and limbic pre- thorax and REST/NRSF to control neural stem cell proliferation and dominant atypical variants of Alzheimer’s disease. Alzheimers differentiation. Cell 2012; 151: 1097–1112. Dement 2012; 8: P160–P161. Zhang X, Mormino EC, Sun N, Sperling RA, Sabuncu MR, Yeo BTT, Winblad B, Palmer K, Kivipelto M, Jelic V, Fratiglioni L, Wahlund et al. Bayesian model reveals latent atrophy factors with dissociable L-O, et al. Mild cognitive impairment–beyond controversies, to- cognitive trajectories in Alzheimer’s disease. Proc Natl Acad Sci USA wards a consensus: report of the International Working Group on 2016; 113: E6535–E6544. Mild Cognitive Impairment. J Intern Med ; 2004; 256: 240–246. Zhang Y, Chen K, Sloan SA, Bennett ML, Scholze AR, O’Keeffe S, et Yamada K, Holth JK, Liao F, Stewart FR, Mahan TE, Jiang H, et al. al. An RNA-sequencing transcriptome and splicing database of glia, Neuronal activity regulates extracellular tau in vivo. J Exp Med neurons, and vascular cells of the cerebral cortex. J Neurosci 2014; 2014; 211: 387–393. 34: 11929–11947.