Konijnenberg et al. Alzheimer's Research & Therapy (2020) 12:65 https://doi.org/10.1186/s13195-020-00628-z RESEARCH Open Access APOE ε4 genotype-dependent cerebrospinal fluid proteomic signatures in Alzheimer’s disease Elles Konijnenberg1, Betty M. Tijms1* , Johan Gobom2,3, Valerija Dobricic4, Isabelle Bos1,5, Stephanie Vos5, Magda Tsolaki6, Frans Verhey5, Julius Popp7, Pablo Martinez-Lage8, Rik Vandenberghe9, Alberto Lleó10, Lutz Frölich11, Simon Lovestone12,13, Johannes Streffer14,15, Lars Bertram4,16,17, Kaj Blennow18,19, Charlotte E. Teunissen20, Robert Veerhuis20,21, August B. Smit22, Philip Scheltens1, Henrik Zetterberg19,20,23,24 and Pieter Jelle Visser1,5,25 Abstract Background: Aggregation of amyloid β into plaques in the brain is one of the earliest pathological events in Alzheimer’s disease (AD). The exact pathophysiology leading to dementia is still uncertain, but the apolipoprotein E (APOE) ε4 genotype plays a major role. We aimed to identify the molecular pathways associated with amyloid β aggregation using cerebrospinal fluid (CSF) proteomics and to study the potential modifying effects of APOE ε4 genotype. Methods: We tested 243 proteins and protein fragments in CSF comparing 193 subjects with AD across the cognitive spectrum (65% APOE ε4 carriers, average age 75 ± 7 years) against 60 controls with normal CSF amyloid β, normal cognition, and no APOE ε4 allele (average age 75 ± 6 years). Results: One hundred twenty-nine proteins (53%) were associated with aggregated amyloid β. APOE ε4 carriers with AD showed altered concentrations of proteins involved in the complement pathway and glycolysis when cognition was normal and lower concentrations of proteins involved in synapse structure and function when cognitive impairment was moderately severe. APOE ε4 non-carriers with AD showed lower expression of proteins involved in synapse structure and function when cognition was normal and lower concentrations of proteins that were associated with complement and other inflammatory processes when cognitive impairment was mild. Repeating analyses for 114 proteins that were available in an independent EMIF-AD MBD dataset (n = 275) showed that 80% of the proteins showed group differences in a similar direction, but overall, 28% effects reached statistical significance (ranging between 6 and 87% depending on the disease stage and genotype), suggesting variable reproducibility. Conclusions: These results imply that AD pathophysiology depends on APOE genotype and that treatment for AD may need to be tailored according to APOE genotype and severity of the cognitive impairment. Keywords: Amyloid aggregation, APOE genotype, CSF proteomics * Correspondence: b.tijms@amsterdamumc.nl 1 Alzheimer Center Amsterdam, Department of Neurology, Amsterdam Neuroscience, Amsterdam UMC, Vrije Universiteit Amsterdam, PO Box 7057, 1007 MB Amsterdam, The Netherlands Full list of author information is available at the end of the article © The Author(s). 2020 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Konijnenberg et al. Alzheimer's Research & Therapy (2020) 12:65 Page 2 of 11 Background Further details about ADNI are given in the “Acknowl- Amyloid β aggregation in the brain is one of the earliest edgments” section. Study protocols were approved by pathological events in Alzheimer’s disease (AD) and is the institutional review boards of all participating ADNI thought to start decades before the manifestation of de- centers (a complete list of ADNI sites is available at mentia [1–3]. The presence of an apolipoprotein E http://adni.loni.usc.edu/about/centers-cores/study-sites/ (APOE) ε4 allele, the major genetic risk factor for AD ), and written informed consent was obtained from all [4], lowers the age of onset through an as of yet un- participants or authorized representatives. All analyses known mechanism. In general, it is largely unclear which were performed on de-identified ADNI data, and biological processes eventually lead to cognitive decline methods were carried out in accordance with the ap- once amyloid β has aggregated, as well as whether such proved guidelines. processes are influenced by the presence of the APOE ε4 For the present study, we included individuals who allele. A better understanding of the biological processes had baseline CSF data available for amyloid β 1–42 and disrupted in AD subjects is crucial for the development proteomics (see the next section). Aggregated amyloid in of precision medicine. The apoE4 protein isoform has CSF was defined as having CSF amyloid β 1–42 levels been associated with impaired amyloid clearance and below 192 pg/ml [16]. APOE genotype was assessed with transport, synaptogenesis, and glucose and cholesterol two SNPs (rs429358, rs7412) that define the epsilon 2, 3, metabolism in the brain [5, 6]. However, about 25–40% and 4 alleles, using DNA extracted by Cogenics from a of patients with AD dementia lack the APOE ε4 allele 3-ml aliquot of EDTA blood. Subjects were classified ac- [7], and for these individuals, the pathophysiological cording to amyloid status (normal/abnormal), APOE ε4 mechanisms involved in AD are less clear [8]. Unbiased genotype (carrier/non-carrier), and cognitive stage as proteomic analysis in cerebrospinal fluid (CSF) allows measured with the MMSE [17] (normal cognition: studying multiple molecular processes at the same time MMSE > 27; mild impairment: MMSE scores between in patients, and it can be hypothesized that distinct pat- 27 and 24; moderate impairment: MMSE < 24). terns of protein concentrations exist in the CSF that are associated with aggregated amyloid. The first CSF prote- CSF protein analysis omic studies have identified novel markers associated CSF samples were collected with a lumbar puncture, and with AD-type dementia when comparing patients with samples were stored at the ADNI Core Laboratory at the cognitively normal controls [9–11]. Yet, not all subjects University of Pennsylvania Medical Center on dry ice with a clinical diagnosis of AD-type dementia have ag- until further analysis. In total, 313 proteins and protein gregated amyloid, and on average, 30% of cognitively fragments were measured: 12 with ELISA, 159 with pro- normal subjects are already in the preclinical stage of teomics RBM, and 142 with proteomics multiple reac- AD [3, 12, 13]. Consequently, it remains uncertain which tion monitoring (MRM) targeted mass spectroscopy (see of the previously reported markers are specific for AD supplementary table 1 for an overview of all included pathology, i.e., aggregated amyloid. Furthermore, protein proteins). Information on protein assessment and quality levels in CSF may depend on APOE ε4 genotype, which control is described at http://adni.loni.usc.edu/data-sam- has been reported for beta secretase-1 [BACE1] [14] and ples/biospecimen-data/. For MRM, we used the finalized chitinase-3-like protein 1 [YKL40] [15], both proteins as- “Normalized Intensity” data [9], which was the result of sociated with AD-type dementia, and so, it is plausible a two-step normalization procedure of the raw peak area that APOE ε4 genotype may influence other protein data to remove variability between samples processed on markers in the CSF as well. different days introduced by the depletion method: First, In this study, we used a CSF proteomic approach to process-related bias was removed by correcting for test the hypothesis that protein signatures can be de- trends when observed, by computing the predicted aver- tected that show APOE ε4 genotype-dependent associa- age log-intensity values from smoothing spline function tions with AD across the cognitive spectrum. to the CSF sample averages. For each sample at a given transition, the predicted value was subtracted from the Methods sample average log-intensity. Second, using two regres- Participants sion models to model the daily sample average and the We downloaded ADNI data in August 2017 from the global sample average, the log-intensity values of the ADNI database (all data is available at adni.loni.usc.edu), CSF samples after step 1 normalized were further including participants from over 50 sites across the USA normalized to account for the depletion day of the sam- and Canada (www.adni-info.org). As such, the investiga- ples. (please see for a detailed explanation of the tors within the ADNI contributed to the design and im- normalization procedure the “Biomarkers Consortium plementation of ADNI and/or provided data but did not CSF ProteomicsMRM data set” in the “Data Primer” participate in the analysis or writing of this report. document at adni.loni.ucla.edu). All CSF protein levels Konijnenberg et al. Alzheimer's Research & Therapy (2020) 12:65 Page 3 of 11 were Z-transformed to the control group (normal amyl- Pathway enrichment analyses oid, APOE ε4 non-carrier, MMSE > 27), such that nega- We used the online database STRING [19] to identify tive values indicate lower and positive values indicate enriched biological processes (based on KEGG pathways higher levels compared to the normal state. If peptides and GO biological processes) for each protein that from the same protein showed a moderate to strong cor- showed significant differences with the control group. In relation (r > .6), we combined peptides into a composite addition, we used this database to test for pathway en- measure by averaging their Z scores. This resulted in richment entering all proteins associated with a particu- 243 protein measures tested. A subset of 114 proteins lar group at the same time. tested in the present study was also measured with tan- dem mass tag spectrometry in the multi-center EMIF- Results AD Multimodal Biomarker Discovery (MBD) study [18] Sample characteristics by APOE ε4 genotype which we used for replication analyses. In total, 253 individuals from the Alzheimer’s Disease Neuroimaging Initiative-1 (ADNI-1) had available base- line proteomic CSF data. Compared to the control Statistical analysis group, APOE ε4 carriers and non-carriers with aggre- t test, χ2, and Kruskal-Wallis tests were used to com- gated amyloid β had similar average age, level of educa- pare the subject characteristics between the AD and tion, and gender distributions (Table 1). control groups. We compared the protein levels be- Figure 1 shows that proteins associated with aggre- tween subjects with AD (defined as having aggregated gated amyloid formed distinct clusters depending on the amyloid) and the control group, stratified for APOE APOE genotype and cognitive stage. In total, 129 (53%) ε4 genotype and cognitive stage with ANCOVAs that proteins and protein fragments were associated with ag- included age and gender as potential confounders. All gregated amyloid, with 27 (21%) proteins showing higher statistical analyses were performed using R, version levels and the majority of proteins (102, 79%) showing 3.2.3. lower levels in AD compared to controls. The large Table 1 Sample description Characteristic APOE ε4 non-carriers APOE ε4 carriers Abeta normal Abeta Abeta Abeta Abeta Abeta Abeta Abeta (control), abnormal, abnormal, abnormal, Normal, Abnormal, Abnormal, Abnormal, MMSE> 27 MMSE > 27 MMSE 27–24 MMSE < 24 MMSE > 27 MMSE > 27 MMSE 27–24 MMSE < 24 N (% total 60 (23%) 24 (9%) 34 (13%) 9 (3%) 8 (3%) 40 (15%) 63 (24%) 23 (27%) sample) E2/E3 15 (25%) 3 (12.5%) 6 (18%) 0 0 0 0 0 E3/E3 45 (75%) 21 (87.5%) 28 (82%) 9 (100%) 0 0 0 0 E2/E4 0 0 0 0 0 1 (2.5%) 2 (3%) 1 (4%) E3/E4 0 0 0 0 8 (100%) 34 (85%) 41 (68%) 13 (57%) E4/E4 0 0 0 0 0 5 (12.5%) 20 (33%) 9 (39%) Age 75.30 (6.44) 76.70 (4.90) 75.37 (8.10) 78.82 (5.55) 72.1 (6.43) 74.97 (6.16) 74.53 (6.88) 73.32 (8.70) Female, N (%) 29 (48%) 10 (42%) 13 (38%) 4 (44%) 4 (50%) 12 (30%) 30 (48%) 8 (35%) Education 16.00 (2.76) 16.08 (3.32) 15.62 (3.29) 16.22 (3.23) 15.50 (2.00) 16.02 (3.22) 15.25 (2.83) 15.30 (2.67) years MMSE 29.15 (0.73) 29.21 (0.78) 25.59 (1.05)c 21.89 (1.05)*** 29.12 (0.64) 28.73 (0.78)** 25.49 (1.11)*** 21.91 (1.12)*** Amyloid β 1– 245.53 (26.38) 150.5 142.29 148.78 235.75 132.05 131.94 128.61 42 pg/ml (28.76)*** (25.06)*** (19.32)*** (22.13) (23.19)*** (28.07)*** (23.52)*** T-tau pg/ml 63.41 (22.05) 88.12 (55.61)* 110.68 149.44 77.12 (19.77) 106.91 120.98 124.91 (51.83)*** (75.91)*** (48.96)*** (54.37)*** (51.75)*** P-tau pg/ml 20.56 (7.86) 31.33 (15.95)* 40.21 (18.04)*** 49.33 22 (4.41) 37.18 39.84 (13.70)*** 44.39 (20.43)*** (14.81)*** (23.78)*** APOE ε4 6.18 (2.18) 5.98 (3.26) 5.84 (2.68) 5.61 (4.44) 10.07 (5.80)c 11.92 (1.24)*** 11.89 (1.43)*** 11.63 (1.17)*** isoform Values are mean (SD). All comparisons are made against the control group (APOE ε4 non-carriers with normal amyloid and MMSE > 27). APOE ε4 isoform is the protein fragment that is specifically encoded by the ε4 allele, average log-intensity value are provided (all non-carriers have values below the level of detection, Spellman et al., 2015) *p < .05, **p < .01, ***p < .001 Konijnenberg et al. Alzheimer's Research & Therapy (2020) 12:65 Page 4 of 11 Fig. 1 Left, heatmap of proteins associated with amyloid pathology. Columns indicate APOE ε4 carriers and non-carriers with AD according to the severity of their cognitive impairment (MMSE > 27, 27–24, or < 24). Color scale indicates the Z value of proteins showing a significant difference (p < .05) compared with the control group (APOE ε4 non-carriers with normal amyloid and MMSE > 27). Proteins are expressed as Z- scores using the control group as a reference and plotted when showing a significant difference (p < .05). Light blue indicates non-significance (p > .05). Right, the percentage of proteins associated with one of the 11 biological process categories. Percentages were calculated with disease stage-specific total numbers of proteins associated with abnormal amyloid, and the number of proteins is given below each disease stage. Please see supplementary table 3 for a detailed description of biological processes enriched majority (90%) of proteins associated with aggregated major pathological hallmark for AD, was the only pro- amyloid showed expression differences with controls tein that showed higher levels in all AD subjects across that depended on the cognitive stage. Tau, another the cognitive spectrum, with higher concentrations for Konijnenberg et al. Alzheimer's Research & Therapy (2020) 12:65 Page 5 of 11 more severe impairment, regardless of APOE ε4 status. enrichment was observed in mild impairment, and for We further observed two patterns of protein expression moderate cognitive impairment “cell adhesion mole- levels: (1) 83 of the 129 proteins (64%) had altered levels cules” (Table 2). either in ε4 carriers or non-carriers; (2) 46 of the 129 proteins (36%) had altered levels in both APOE ε4 car- riers and non-carriers, but in different cognitive stages. APOE ε4 non-carrier associations of proteins with aggregated amyloid β APOE ε4 genotype associations of proteins with APOE ε4 non-carriers with aggregated amyloid showed aggregated amyloid β a different proteomic profile than APOE ε4 carriers, in Compared to controls, APOE ε4 carriers with normal the type of proteins expressed and/or the cognitive stage cognition showed higher levels of nicastrin [NCSTN], of expression. Non-carriers with normal cognition which is part of the gamma secretase complex, and of a showed lower levels of a large group of proteins associ- group of proteins that were associated with glycolysis ated with cell adhesion processes compared to the (Fig. 1, first column; supplementary table 1). Carriers control group (Fig. 1; supplementary table 2). A subset further showed higher levels of markers known to in- of these proteins included synaptic markers contactin-1 crease with neuronal injury (neurogranin [NRGN]; fatty [CNTN1], neurexin-1 [NRXN1] and neurexin-2 acid-binding protein, heart [FABP3]; visinin-like protein- [NRXN2] that were associated with “transsynaptic sig- 1[VILIP1]; YKL40), growth factors fibroblast growth naling” [21], and the neuronal pentraxin receptor factor-4 [FGF4], and hepatocyte growth factor [HGF] [NPTXR]. In this stage, only tau showed higher levels. [20]. These proteins were also higher in subjects with APOE ε4 non-carriers with mild impairment showed mild and moderate cognitive impairment. Furthermore, lower levels of complement-related proteins, which over- a large group of proteins had lower levels in AD, includ- lapped with the complement proteins that showed lower ing immune system-related complement factors (C2, C3, levels in APOE ε4 carriers with normal cognition. Fur- C5, C6, factor-B [CFB], factor-H [CFH]) and factors that ther, alterations observed in non-carriers with mild im- interact with the complement system (plasminogen pairment were higher levels of a wide range of neuronal [PLG], prothrombin [F2], serum amyloid P component injury markers (NRGN, FAPB3, VILIP1). APOE ε4 non- [APCS], and C-reactive protein [CRP]). Subjects with carriers with moderate cognitive impairment also had moderate cognitive impairment showed lower levels of higher levels of glial fibrillary acidic protein [GFAP], proteins that were mostly associated with cell adhesion- neurofilament light [NFL], resistin [RETN], and macro- related processes (Fig. 1, fifth column), including phage migration inhibitory factor [MIF]. These proteins markers functionally associated with “transsynaptic sig- did not show a clear association with a shared biological naling”(e.g., cadherin-13 [CDH13], neogenin [NEO1], pathway (Fig. 1, sixth column), but might be related to neural cell adhesion molecule-1 [NCAM1], neuronal cell inflammatory responses. In addition, sortilin [SORT1] adhesion molecule [NRCAM]), “peptide neurotrophin levels were higher in these subjects. SORT1 has several signaling” (chromogranin-A [CHGA], proenkephalin-B functions and is involved in APP processing [22]. No [PDYN], secretogranin-2 [SCG2], proSAAS [PCSK1]), proteins showed lower levels in this stage, but it should and “GPCR signaling”(glutamate receptor-4 [GRM4], be noted that this group had a small sample size, which latrophilin-1 [ADGRL1]) [21]. The top pathways may have limited statistical power. The top pathways enriched in KEGG for proteins associated with aggre- enriched in KEGG for proteins associated with aggre- gated amyloid were “complement and coagulation gated amyloid in APOE ε4 non-carriers were “cell adhe- cascades” for subjects with normal cognition; no sion molecules” for subjects with normal cognition and Table 2 Summary of pathways enriched in KEGG of proteins associated with aggregated amyloid according to APOE ε4 genotype Cognitive APOE ε4 carriers APOE ε4 non-carriers stage Pathway enriched pFDR Proteins Pathway enriched pFDR Proteins MMSE > Complement and 8.20E C2, C3, C5, C6, CFB, CFH, F2, KNG1, PLG Cell adhesion 0.000165 CADM3’, NRXN2, 27 coagulation cascades −12 molecules NRXN1, CNTN1, CNTN AP2 MMSE No enrichment n.a. n.a. Complement and 8.61E C2, C3, C5, C6, CFB, F2, 27–24 detected coagulation cascades −13 KNG1, PLG MMSE < Cell adhesion 8.44E CADM3’, CNTN1, CNTNAP2, NCAM1, NEGR1, No enrichment n.a. n.a. 24 molecules −09 NEO1, NRCAM, NRXN1, NRXN2, PVRL1’ detected Please note that APOE and APP were excluded from enrichment analyses. Bold font indicates concordant difference compared to controls observed in EMIF-AD MBD; apostrophe (’) indicates not available in independent cohort n.a not applicable, MMSE Mini-Mental State Examination, FDR false discovery rate Konijnenberg et al. Alzheimer's Research & Therapy (2020) 12:65 Page 6 of 11 “complement and coagulation cascades” for subjects aggregation in the carrier or non-carrier groups, 114 with mild impairment (Table 2). were also measured in the EMIF-AD MBD, and across all contrasts, 80% showed concordant group differences APOE ε4 effect on amyloid processing in asymptomatic compared to controls. Within in the APOE ε4 carriers, subjects with normal amyloid β individuals with mild and moderate cognitive impair- We further explored whether protein differences could ment showed high concordances in group differences of be observed in APOE ε4 carriers with normal amyloid β respectively 93% of 13 overlapping proteins and 95% of and cognition, as these subjects are at increased genetic 63 overlapping proteins (Fig. 3; supplementary table 1), risk of developing amyloid pathology [4], and so, for and individuals with normal cognition showed moderate these subjects, proteomic alterations may indicate very concordance of 58% of 40 overlapping proteins. APOE early pathological changes associated with AD. Injury ε4 carriers with normal cognition and normal amyloid markers were normal in these subjects. Compared to the showed concordance of 100% of 4 overlapping proteins. control group, ten proteins (APOE ε4 fragment, tau In the APOE ε4 non-carriers, individuals with mild im- [MAPT], BACE1, β-nerve growth factor [NGF], macro- pairment showed a low level of concordance of 28% of phage inflammatory protein-1β [CCL4], osteopontin 36 overlapping proteins, and individuals with normal [SPP1], AXL receptor tyrosine kinase [AXL], heparin- cognition and moderate cognitive impairment showed binding EGF-like growth factor [HBGF], carbonic high concordance (respectively 87% of 15 overlapping anhydrase-1 [CA1], interferon gamma-induced protein- proteins and 100% of 7 overlapping proteins; Fig. 3; sup- 10 [CXCL10]) showed altered levels (Fig. 2). The top plementary table 2). KEGG pathway enriched was the Toll-like receptor signaling pathway (pFDR = 0.00033; including SPP1, Discussion CXCL10, and CCL4). Summary In this study, both APOE ε4 carriers and non-carriers Replication in the independent EMIF-AD MBD cohort with abnormal CSF amyloid β 1–42 levels showed alter- Finally, we performed replication analyses in the inde- ations of large groups of proteins involved in neuronal pendent EMIF-AD MBD cohort by comparing the con- injury, complement and inflammatory processes, and cell cordance of observed group differences compared to adhesion processes, but in a different temporal ordering. controls. Overall, individuals in EMIF-AD MBD were APOE ε4 carriers showed altered protein levels of com- younger than in ADNI and showed lower levels of edu- plement related proteins in the normal cognition stage, cation (supplementary table 4). APOE ε4 carriers with while lower levels of proteins associated with cell adhe- mild impairment were more often female in EMIF-AD sion and synaptic signaling were found in cognitive im- MBD (54%) than in ADNI (30%). Of the 136 proteins/ pairment stages. Non-carriers with aggregated amyloid fragments that showed an association with amyloid showed a reversed temporal ordering of these processes Fig. 2 Proteins associated with APOE ε4 carrier status in subjects with normal amyloid. Z-scores are plotted for proteins that were different between subjects with APOE ε4 (in brown) and normal amyloid and normal cognition (MMSE > 27) compared to the control group (i.e., in blue). All values are standardized according to the control group (i.e., APOE ε4 non-carriers with normal amyloid and MMSE > 27) Konijnenberg et al. Alzheimer's Research & Therapy (2020) 12:65 Page 7 of 11 a) APOE e4 carriers MMSE>27 versus controls APOE e4 carriers MMSE 27-24 versus controls APOE e4 carriers MMSE <24 versus controls APOE e4 carriers MMSE >27 with normal amayloid versus controls SERPINA1_AVLTIDEK BTD APOA1 SPARCL1 C1QB EXTL2 CHGA_EDSLEAGLPLQVR CUTA NCAM1 CLU SERPINF1 TNFRSF21 CLSTN1 SORCS1 DAG1 NEO1 Abeta 1−40 CD14 CDH13 CHGB NECTIN1 B4GAT1 PLXDC1 PTGDS OGN FAM3C NELL2 Abeta 1−38 CNTN1 SCG2 Protein NRXN1 NEGR1 A1BG CADM3 CFB FBLN1 AHSG NRXN3 MB PLG NCAN SERPINA7 SEZ6L CP CD59 GC C3_TGLQEVEVK NRCAM F2 NBL1 APOA1 SCG3 C6 HPX ITIH5 ITIH1 ADGRL1 LRG1 CCK PGLYRP2 APP_LVFFAEDVGSNK AMBP C5 PCMT1 Protein CST3 NRXN2 TIMP1 PTPRN OGN C3 PDYN APOH MCAM APCS CACNA2D1 Protein C2 CFH MCAM NPTX1 SERPINA4 APOA1 PAM KNG1 NPTXR PCSK1 AFM NPTX2 VGF CHI3L1 APOB_IAELSATAQEIIK GOT2 NPTX2 NCSTN CHI3L1 NPTXR CA1 GOT1 ENO2_GNPTVEVDLYTAK APOB_IAELSATAQEIIK PKM S100B Protein FABP3 ALDOA SPP1 SPP1 ALDOA SPP1 CHI3L1 ENO2_GNPTVEVDLYTAK FABP3 FABP3 MAPT SPP1 NRGN NRGN NRGN MAPT MAPT MAPT AXL −5.0−4.5−4.0−3.5−3.0−2.5−2.0−1.5−1.0−0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 −5.0−4.5−4.0−3.5−3.0−2.5−2.0−1.5−1.0−0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 −5.0−4.5−4.0−3.5−3.0−2.5−2.0−1.5−1.0−0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 −5.0−4.5−4.0−3.5−3.0−2.5−2.0−1.5−1.0−0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 Z score Z score Z score Z score ADNI EMIF ADNI EMIF ADNI EMIF ADNI EMIF b) APOE e4 non-carriers MMSE>27 versus controls APOE e4 non-carriers MMSE 27-24 versus controls APOE e4 non-carriers MMSE <24 versus controls CNTNAP2_VDNAPDQQNSHPDLAQEEIR SERPINF1 NPTX2 NPTXR C2 SORCS1 MCAM PCSK1 CFB F2 A1BG C6 AHSG PTPRN CADM3 PLG PAM VGF CP BTD AFM EFEMP1 GC CD14 FMOD PCSK1 EXTL2 HPX COCH Protein Protein DAG1 NPTXR CLU NRXN1 C3 CNTN1 OGN NPTX1 KNG1 CLU SERPINA4 MIF Protein NRXN2 C5 SPP1 FABP3 SORCS1 SORT1 FABP3 SPP1 NPTX2 APOD_VLNQELR NEFL CNTNAP2_VDNAPDQQNSHPDLAQEEIR NRGN NRGN MAPT MAPT MAPT −5.0−4.5−4.0−3.5−3.0−2.5−2.0−1.5−1.0−0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 −5.0−4.5−4.0−3.5−3.0−2.5−2.0−1.5−1.0−0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 −5.0−4.5−4.0−3.5−3.0−2.5−2.0−1.5−1.0−0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 Z score Z score Z score ADNI EMIF ADNI EMIF ADNI EMIF Fig. 3 Effect size plots for 114 proteins measured in both ADNI and EMIF-AD MBD studies. a Effects for protein differences of individuals with abnormal amyloid carrying and ≥ 1 APOE ε4 allele against controls (i.e., APOE ε4 non-carriers with normal amyloid and MMSE > 27) according to the severity of their cognitive impairment (MMSE > 27, 27–24, or < 24), and at the far left for APOE ε4 carriers with normal amyloid and MMSE > 27. In blue, effect sizes and 95%CI for ADNI; in orange, for EMIF-AD MBD. b Effects for protein differences of individuals with abnormal amyloid carrying without an APOE ε4 allele against controls (i.e., APOE ε4 non-carriers with normal amyloid and MMSE > 27) according to the severity of their cognitive impairment (MMSE > 27, 27–24, or < 24). In blue, effect sizes and 95%CI for ADNI; in orange, for EMIF-AD MBD with proteins involved in cell adhesion processes show- carriers showed alterations in the mild impairment stage. ing altered levels in cognitively normal subjects, which Possibly, this observation reflects that the apoE4 protein was followed by alterations in complement-related pro- enhances complement activation in the presence of ag- teins in cognitive impairment stages. These results sug- gregated amyloid β [29]. However, only a subset of these gest that subjects with AD may require specific proteins (KNG1, CFH, C2) showed similar differences in treatment tailored to their APOE genotype and degree the replication cohort, and these were less pronounced. of cognitive impairment. The replication cohort further showed contrary group differences compared to ADNI for C6, CFB, F2, and CSF proteome signatures associated with APOE ε4 PLG in both APOE carriers with normal cognition and APOE ε4 carriers with normal cognition showed lower in APOE non-carriers with mild impairment. This sug- levels of complement-related proteins C2, C3, C5, C6, gests that the different proteomic techniques may differ CFB, CFH, PLG, F2, APCS, and CRP. The complement in sensitivity to measure these proteins accurately, and system is a major part of the innate immune system, and future research should further investigate this issue by its classical activation pathway can be directly triggered comparing both techniques within the same cohort. At by amyloid aggregates [23, 24]. Previous studies investi- this point, we can only speculate as to how reduced gating complement-related protein concentrations in levels of complement proteins in CSF can be interpreted. CSF have, however, reported divergent results with Lower protein concentrations may reflect binding of higher concentrations in AD-type dementia patients complement proteins to pathogen surfaces, possibly to [25–27] and also lower concentrations in AD-type de- tag these for phagocytosis [23], and the presence of com- mentia patients [28] and in subjects with mild cognitive plement proteins in amyloid plaques seems to support impairment who showed cognitive decline at follow-up this explanation [23, 24, 30]. Alternatively, lower levels [27]. Our results suggest that levels may be altered in of complement proteins could point towards decreased different cognitive stages according to APOE genotype, production, which seems to be in line with the observa- with ε4 carriers showing more extensive complement in- tion that regulators of complement activation like CFH volvement in the cognitively normal stage, whereas non- also showed lower levels in these subjects. Furthermore, Konijnenberg et al. Alzheimer's Research & Therapy (2020) 12:65 Page 8 of 11 carriers in the moderate cognitive impairment stage APOE ε4 carriers showed lower levels of a small group showed lower levels than controls of complement C1q of cell adhesion molecules, and substantially more cell subcomponent subunit-B [C1QB] and CD59, which was adhesion proteins showed lower levels in the moderate also observed in the replication cohort. C1q can directly impairment stage, among which proteins associated with bind to amyloid β fibrils, which can lead to activation of synapse development (NPTXR, NRCAM, NEO1, C1 as well as C3 [24, 31]. Whereas C3 is associated with NCAM1, CDH13), which was also observed in the repli- several pathways of the complement system, C1QB is cation cohort [21]. In addition, proteins that showed specific for classical complement activation [23, 24]. The lower concentrations than controls were associated with involvement of different complement proteins according presynaptic dense core vesicles (CHGA, secretogranin-3 to cognitive stage suggests that triggers of the comple- [SGC3], voltage-dependent calcium channel subunit ment system might exist that depend on the level of alpha-2 delta-1 [CACNA2D1], PDYN, CDH13, SPARC- neuronal injury and/or the degree of amyloid fibril for- like protein 1[SPARCL1], alpha-1 antitrypsin [SER- mation. Future research should further study how com- PINA1]) [45]. These proteins are associated with peptide plement levels change longitudinally during the neurotrophin signaling. Since synapse loss correlates development of AD, and how these processes depend on well with cognitive decline [46], and in more severe cog- APOE genotype. nitive impairment stages these proteins showed lower Alterations of complement protein concentrations levels, also in the replication cohort, this might indicate were accompanied by a range of inflammatory markers impaired synaptic functioning. However, APOE ε4 non- in APOE ε4 carriers, some of these showing altered carriers showed normal levels of the majority of these levels in carriers who had still normal amyloid β levels, proteins despite the same stage of cognitive impairment suggesting that inflammation processes may play a role and similar levels of neuronal injury markers. in the development in AD before amyloid aggregation becomes manifest in CSF. Some of these markers have CSF proteome signatures associated with aggregated been associated with microglia dysfunction or response amyloid β in APOE ε4 non-carriers associated with neurodegeneration (AXL, SPP1, FABP3) Non-carriers with aggregated amyloid and normal cogni- [32, 33] and reactive astrocytes (CCL4, S100 calcium tion showed lower levels of presynaptic proteins (NRX1 binding protein B [S100B], YKL40, GFAP) [34]. In and NRX2), proteins involved intracellular trafficking APOE ε4 non-carriers, most of these protein levels were (vacuolar protein sorting 10 [VPS10], domain-containing similar to controls, except for inflammation markers receptor SorCS1 [SORCS1]), and neuronal pentraxins SPP1, FABP3 (also in the replication cohort), and GFAP [47], suggesting alterations in the presynaptic cell struc- that were higher in more severe stages of cognitive im- ture may be an early event in AD for subjects lacking pairment. Together, these results support the notion that the ε4 allele. These differences with controls were also inflammation plays an important role in AD [35], and observed in the replication cohort, albeit less pro- we further show that the timing of involved inflamma- nounced. In particular, SORCS1 stands out in this tion processes seem to depend on APOE genotype. It is group, as this gene has been associated with an increased conceivable that these differences reflect apoE isoform- risk for AD [48], and this protein plays a key role in specific interactions with astrocyte and microglia func- intracellular sorting and trafficking of proteins, including tioning [32, 33, 36, 37]. APP, neuronal pentraxins, and NRX1 and NRX2 [48– APOE ε4 carriers with normal amyloid β showed 50]. This leads us to propose that the lower levels we higher levels of BACE1, which is the secretase that initi- observed presently in ε4 non-carriers with still normal ates amyloidogenic processing of APP [38]. This suggests cognition might reflect disruption of these cellular trans- that increased APP processing might be a pre-amyloid port mechanisms and subsequent failure of intracellular event [39]. Tau levels were also already increased, al- processes such as protein recycling, exocytosis, or autop- though lower than in carriers with normal cognition hagocytosis. Levels of the synaptic proteins were low in who showed aggregated amyloid. Cognitively normal the mild impairment stage, and in that stage, additional APOE ε4 carriers with aggregated amyloid further proteins associated with cell adhesion processes, like, showed higher levels of proteins associated with glycoly- e.g., cell adhesion molecule-3 [CADM3] also showed sis, which was also observed in the replication cohort. lower levels. The replication cohort showed, however, High levels of proteins involved in glycolysis have previ- higher levels for those proteins. It is unclear whether this ously been reported in brain pathology studies in early reflects differences in disease stage, or age, as the repli- stage AD [40]. APOE ε4 has been associated to dysfunc- cation cohort was younger. Differences in the results be- tion of mitochondria [41], and so increased levels of gly- tween the cohorts may also reflect the differences colysis may indicate compensation for mitochondrial between proteomic techniques. More research is needed dysfunction [41–44]. In the mild impairment stage, that use the same proteomics technique in larger Konijnenberg et al. Alzheimer's Research & Therapy (2020) 12:65 Page 9 of 11 samples of non-carriers with mild impairment to further Furthermore, we have operationalized disease severity investigate this issue. Another finding was that we ob- in our sample based on the MMSE, which is a screening served higher NFL levels in non-carriers in both cohorts. tool. An alternative approach would have been Higher levels of CSF NFL indicate axonal injury, and categorization based on syndrome diagnosis, but a draw- such higher levels have been associated with neurode- back of that approach is that individuals with normal generative processes in several neurological disorders [51]. cognition, MCI, and dementia can have the same MMSE. At this point, no tools exist that accurately de- Strengths and limitations lineates disease severity in a non-demented population, A potential limitation of this study is that the interpret- and future research should focus on developing tools ation of higher and lower levels of proteins measured in that are sensitive to cognitive impairment in pre- CSF in terms of activated biological processes is not al- dementia stages of AD. Another potential limitation is ways straightforward. However, interpretations for some that we labeled proteins based on the top pathways proteins, such as amyloid β and tau, have been well found, and although this simplifies the interpretation of established through associations with histopathological the results, this approach does not take into account the measurements in post-mortem research [52] and/or with notion that proteins could be involved in multiple bio- PET imaging [53]. Still, 80% of the associations for pro- logical processes. In addition, in the present study, we teins that were also measured in an independent repli- defined AD based on abnormal CSF amyloid levels, be- cation cohort showed group differences in a similar cause (as of yet) the majority (97%) of subjects did not direction, but this depended on the specific group stud- have pathological data available, which can be seen as a ied, and small sample sizes of these subgroups (particu- limitation of this study. Still, it has been shown previ- larly so for the non-carriers with MMSE < 24 and the ously that amyloid biomarker values correlate with histo- carriers with normal amyloid and MMSE > 27) may have pathological measures for amyloid plaques [52]. Using limited the statistical power to observe significant re- biomarkers to define AD can also be regarded to be a sults. Therefore, more studies are needed to replicate strong aspect of our study. This way, we avoided poten- these findings in larger samples. Nevertheless, our re- tial biases in our results that may arise when defining sults may be useful to select proteins for further detailed groups based only on clinical characteristics, as clinical studies, as these seem to be involved in AD pathology features do not always accurately reflect the underlying in vivo. Another point of note is that proteins measured pathology. Finally, although our results suggest that sev- in ADNI were specifically selected based on their associ- eral processes associated with aggregated amyloid might ation with biological processes relevant for AD. As such, be transient, further longitudinal CSF proteomic studies pathway enrichment analyses should not be seen as dis- are required to investigate these dynamics in more covery of new processes involved, but rather were meant detail. to aid interpretation of the results. The enriched pro- cesses associated with abnormal amyloid where specific Conclusions for clinical stage and genotype, and it is unlikely that en- In conclusion, we found CSF proteomic signatures that richment of these processes simply reflect the processes were associated with aggregated amyloid β and were associated with the proteins that were selected in this dependent on APOE ε4 genotype and cognitive stage. study, because the background set of proteins was the An implication of our results is that AD subjects may re- same each analysis. Untargeted proteomic techniques quire treatments tailored to APOE genotype and that should be used in order to discover involvement of po- clinical trials may need to consider APOE ε4 dependent tential other pathways that may be associated with AD endpoints in CSF. in an APOE dependent way. Another point of note is that the control group contained more E2 carriers than Supplementary information the APOE ε4 carriers, and consequently, differences Supplementary information accompanies this paper at https://doi.org/10. between these groups may also contain E2-specific 1186/s13195-020-00628-z. effects, which may not directly reflect amyloid abnormal- Additional file 1. Supplementary tables. ity. Furthermore, APOE ε4 carriers with MMSE > 27 were less often ε4 homozygous than carriers with MMSE < 27 and abnormal amyloid, and such dose effects may Abbreviations CSF : Cerebrospinal fluid; APOE: Apolipoprotein E; AD: Alzheimer’s disease; have contributed to the different sets of proteins MMSE: Mini Mental-State Examination; BACE1: Beta secretase 1; observed to be related with abnormal amyloid between YKL40: Chitinase-3-like protein 1; ADNI: Alzheimer’s Disease Neuroimaging these stages. Future studies should aim to collect larger Initiative; RBM: Rules-based medicine; MRM: Multiple reaction monitoring; NCSTN: Nicastrin; NRGN: Neurogranin; FABP3: Fatty acid-binding protein, samples of individuals in order to determine dosage heart; VILIP1: Visinin-like protein-1; FGF4: Fibroblast growth factor; effects on the CSF proteome. HGF: Hepatocyte growth factor; C2, C3, C5, C6: Complement factors 2, 3, 5, 6; Konijnenberg et al. Alzheimer's Research & Therapy (2020) 12:65 Page 10 of 11 CFB: Complement factor B; CFH: Complement factor H; PLG: Plasminogen; Competing interests F2: Prothrombin; APCS: Serum amyloid P component; CRP: C-reactive All authors report no potential conflict of interest with the content of this protein; CDH13: Cadherin-13; NEO1: Neogenin; NCAM1: Neural cell adhesion paper. molecule-1; CHGA: Chromogranin-A; PDYN: Proenkephalin; SCG2: Secretogranin-2; PCSJ1: proSAAS; GRM4: Glutamate receptor-4; Author details 1 ADGRL1: Latrophilin-1; CNTN1: Contactin-1; NRXN1: Neurexin-1; Alzheimer Center Amsterdam, Department of Neurology, Amsterdam NRXN2: Neurexin-2; NPTXR: Neuronal pentraxin receptor; GFAP: Glial fibrillary Neuroscience, Amsterdam UMC, Vrije Universiteit Amsterdam, PO Box 7057, acidic protein; NFL: Neurofilament light; RETN: Resistin; MIF: Macrophage 1007 MB Amsterdam, The Netherlands. 2Clinical Neurochemistry Laboratory, migration inhibitory factor; SORT1: Sortilin; NGF: β-Nerve growth factor; Sahlgrenska University Hospital, Mölndal, Sweden. 3Department of Psychiatry CCL4: Macrophage inflammatory protein-1β; SPP1: Osteopontin; AXL: AXL and Neurochemistry, Institute of Neuroscience and Physiology, Sahlgrenska Receptor tyrosine kinase; CLSTN1: Calsyntenin-1; CXCL9: Monokine induced Academy at the University of Gothenburg, Mölndal, Sweden. 4Lübeck by gamma interferon; CA1: Carbonic anhydrase-1; CXCL10: Interferon Interdisciplinary Platform for Genome Analytics, University of Lübeck, Lübeck, gamma-induced protein-10; S100B: S100 calcium-binding protein B; Germany. 5Department of Psychiatry and Neuropsychology, School for SPARCL1: SPARC-like protein 1; SERPINA1: Alpha-1 antitrypsin; Mental Health and Neuroscience, Alzheimer Centre Limburg, Maastricht VSP10: Vacuolar protein sorting 10; SORCS1: Domain-containing receptor University, Maastricht, The Netherlands. 61st Department of Neurology, SorcCS1; CADM3: Cell adhesion molecule-3 AHEPA University Hospital, Thessaloniki, Macedonia, Greece. 7Department of Psychiatry, University Hospital of Lausanne, Lausanne, Switzerland. 8 Department of Neurology, Center for Research and Advanced Therapies, Acknowledgements CITA-Alzheimer Foundation, San Sebastian, Spain. 9University Hospital ADNI is funded by the National Institute on Aging, the National Institute of Leuven, Leuven, Belgium. 10Hospital de la Santa Creu i Sant Pau, Barcelona, Biomedical Imaging and Bioengineering, and through generous Spain. 11Department of Geriatric Psychiatry, Zentralinstitut für Seelische contributions from the following: AbbVie, Alzheimer’s Association; Gesundheit, University of Heidelberg, Mannheim, Germany. 12Department of Alzheimer’s Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Psychiatry, University of Oxford, Oxford, UK. 13Janssen R&D, Beerse, Belgium. Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Eisai Inc.; Elan Pharma- 14 Early Clinical Neurology, UCB Biopharma SPRL, Braine-l’Alleud, Belgium. ceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd. 15 Present Address: Janssen R&D, LLC, Beerse, Belgium. 16School of Public and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Health, Imperial College London, London, UK. 17Department of Psychology, Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; University of Oslo, Oslo, Norway. 18Clinical Neurochemistry Lab, Institute of Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity; Neuroscience and Physiology, Sahlgrenska University Hospital, Mölndal, Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Sweden. 19Department of Psychiatry and Neurochemistry, Institute of Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Neuroscience and Physiology, University of Gothenburg, Mölndal, Sweden. Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition 20 Neurochemistry Laboratory and Biobank, Department of Clinical Chemistry, Therapeutics. The Canadian Institutes of Health Research is providing funds Amsterdam Neuroscience, Amsterdam UMC, Vrije Universiteit Amsterdam, to support ADNI clinical sites in Canada. Private sector contributions are facil- Amsterdam, The Netherlands. 21Department of Psychiatry, Amsterdam itated by the Foundation for the National Institutes of Health (www.fnih.org). Neuroscience, Amsterdam UMC, Vrije Universiteit Amsterdam, Amsterdam, The grantee organization is the Northern California Institute for Research and The Netherlands. 22Department of Molecular and Cellular Neurobiology, Education, and the study is coordinated by the Alzheimer’s Disease Coopera- Center for Neurogenomics and Cognitive Research, Amsterdam tive Study at the University of California, San Diego. ADNI data are dissemi- Neuroscience, VU University Amsterdam, Amsterdam, The Netherlands. nated by the Laboratory for Neuroimaging at the University of Southern 23 Department of Molecular Neuroscience, UCL Institute of Neurology, California. London, UK. 24UK Dementia Research Institute, London, UK. 25Department of ADNI: Data used in the preparation of this article were obtained from the Neurobiology, Care Sciences and Society, Division of Neurogeriatrics, Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc. Karolinska Instutet, Stockholm, Sweden. 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 Received: 18 October 2019 Accepted: 6 May 2020 the 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. References Authors’ contributions 1. Jack CR Jr, Wiste HJ, Weigand SD, Rocca WA. Age-specific population EK and BMT performed the analyses. PJV and BMT contributed to the study frequencies of cerebral β-amyloidosis and neurodegeneration among design. EK, BMT, PJV, PS, AS, RV, and CT drafted the paper. The authors read people with normal cognitive function aged 50–89 years: a cross-sectional and approved the final manuscript. study. Lancet Neurol. 2014;14:997–1005. 2. Bateman RJ, et al. Clinical and biomarker changes in dominantly inherited Funding Alzheimer’s disease. N Engl Med. 2012;367:795–804. ZonMW Memorabel Grant no. 733050824. EU/EFPIA Innovative Medicines 3. Jansen WJ, et al. Prevalence of cerebral amyloid pathology in persons Initiative Joint Undertaking EMIF grant agreement no. 115372. without dementia. JAMA. 2015;313:1924. 4. Corder EH, et al. Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families. Science. 1993;261:921–3. Availability of data and materials Data collection and sharing for this project was funded by the Alzheimer’s 5. Rohn T. Proteolytic cleavage of apolipoprotein E4 as the keystone for the Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant heightened risk associated with Alzheimer’s disease. IJMS. 2013;14:14908–22. U01 AG024904) and DOD ADNI (Department of Defense award number 6. Lee LC, Goh MQL, Koo EH. Transcriptional regulation of APP by apoE: W81XWH-12-2-0012). to boldly go where no isoform has gone before. BioEssays. 2017;39: 1700062–9. 7. Ward A, et al. Prevalence of apolipoprotein E4 genotype and homozygotes Ethics approval and consent to participate (APOE e4/4) among patients diagnosed with Alzheimer’s disease: a Study protocols were approved by the institutional review boards of all systematic review and meta-analysis. Neuroepidemiology. 2012;38:1–17. participating ADNI centers (a complete list of ADNI sites is available at http:// 8. Jun G, et al. A novel Alzheimer disease locus located near the gene adni.loni.usc.edu/about/centers-cores/study-sites/), and written informed encoding tau protein. Mol Psychiatry. 2015;21:108–17. consent was obtained from all participants or authorized representatives. 9. Spellman DS, et al. Development and evaluation of a multiplexed mass spectrometry based assay for measuring candidate peptide biomarkers in Consent for publication Alzheimer’s Disease Neuroimaging Initiative (ADNI) CSF. Prot Clin Appl. Not applicable. 2015;9:715–31. Konijnenberg et al. Alzheimer's Research & Therapy (2020) 12:65 Page 11 of 11 10. Perrin RJ, et al. Identification and validation of novel cerebrospinal fluid 37. Chung W-S, et al. Novel allele-dependent role for APOE in controlling the biomarkers for staging early Alzheimer’s disease. PLoS One. 2011;6:e16032– rate of synapse pruning by astrocytes. Proc Natl Acad Sci U S A. 2016;113: 23. 10186–91. 11. Llano DA, Bundela S, Mudar RA, Devanarayan V, for the Alzheimer’s Disease 38. Vassar R. BACE1: the beta-secretase enzyme in Alzheimer’s disease. J Mol Neuroimaging Initiative (ADNI). A multivariate predictive modeling Neurosci. 2004;23:105–14. approach reveals a novel CSF peptide signature for both Alzheimer’s 39. Tijms BM, et al. Pre-amyloid stage of Alzheimer’s disease in cognitively disease state classification and for predicting future disease progression. normal individuals. Ann Clin Transl Neurol. 2018;126:659–11. PLoS One. 2017;12:e0182098–18. 40. Hondius DC, et al. Profiling the human hippocampal proteome at all 12. Ossenkoppele R, et al. Prevalence of amyloid PET positivity in dementia pathologic stages of Alzheimer’s disease. Alzheimers Dement. 2016;12:654– syndromes. JAMA. 2015;313:1939. 68. 13. Jellinger KA, Attems J. Prevalence of dementia disorders in the oldest-old: 41. Valla J, et al. Reduced posterior cingulate mitochondrial activity in expired an autopsy study. Acta Neuropathol. 2010;119:421–33. young adult carriers of the APOE ε4 allele, the major late-onset Alzheimer’s 14. Ewers M, et al. Increased CSF-BACE 1 activity is associated with ApoE-ε4 susceptibility gene. J Alzheimers Dis. 2010;22:307–13. genotype in subjects with mild cognitive impairment and Alzheimer’s 42. Perkins M, et al. Altered energy metabolism pathways in the posterior disease. Brain. 2007;131:1252–8. cingulate in young adult apolipoprotein E ɛ4 carriers. J Alzheimers Dis. 2016; 15. Gispert JD, et al. The APOE ε4 genotype modulates CSF YKL-40 levels and 53:95–106. their structural brain correlates in the continuum of Alzheimer’s disease but 43. Nakamura T, Watanabe A, Fujino T, Hosono T, Michikawa M. Apolipoprotein not those of sTREM2. Alzheimers Dement. 2017;6:50–9. E4 (1–272) fragment is associated with mitochondrial proteins and affects 16. Shaw LM, et al. Cerebrospinal fluid biomarker signature in Alzheimer’s mitochondrial function in neuronal cells. Mol Neurodegener. 2009;4:35–11. disease neuroimaging initiative subjects. Ann Neurol. 2009;65:403–13. 44. Chang S, et al. Lipid- and receptor-binding regions of apolipoprotein E4 17. Folstein MF, Folstein SE, McHugh PR. ‘Mini-mental state’. A practical method fragments act in concert to cause mitochondrial dysfunction and for grading the cognitive state of patients for the clinician. J Psychiatr Res. neurotoxicity. Proc Natl Acad Sci U S A. 2005;102:18694–9. 1975;12:189–98. 45. Wegrzyn JL, et al. Proteomics of dense core secretory vesicles reveal distinct 18. Bos I, et al. The EMIF-AD Multimodal Biomarker Discovery study: design, protein categories for secretion of neuroeffectors for cell−cell methods and cohort characteristics. Alzheimers Res Ther. 2018;10:64. communication. J Proteome Res. 2010;9:5002–24. 19. Szklarczyk D, et al. The STRING database in 2017: quality-controlled protein– 46. Selkoe DJ. Alzheimer’s disease is a synaptic failure. Science. 2002;298:789–91. protein association networks, made broadly accessible. Nucleic Acids Res. 47. Dean C, Dresbach T. Neuroligins and neurexins: linking cell adhesion, 2017;45:D362–8. synapse formation and cognitive function. Trends Neurosci. 2006;29:21–9. 20. Fagan AM, Perrin RJ. Upcoming candidate cerebrospinal fluid biomarkers of 48. Hermey G. The Vps10p-domain receptor family. Cell Mol Life Sci. 2009;66: Alzheimer’s disease. Biomarkers Med. 2012;6:455–76. 2677–89. 21. Ruano D, et al. Functional gene group analysis reveals a role of synaptic 49. Reitz C, et al. SORCS1 alters amyloid precursor protein processing and heterotrimeric G proteins in cognitive ability. Am J Hum Genet. 2010;86: variants may increase Alzheimer’s disease risk. Ann Neurol. 2011;69:47–64. 113–25. 50. Lane RF, et al. Protein sorting motifs in the cytoplasmic tail of SorCS1 22. Gustafsen C, et al. Sortilin and SorLA display distinct roles in processing and control generation of Alzheimer’s amyloid-β peptide. J Neurosci. 2013;33: trafficking of amyloid precursor protein. J Neurosci. 2013;33:64–71. 7099–107. 51. Zetterberg H, et al. Association of cerebrospinal fluid neurofilament light 23. Akiyama H. Inflammation and Alzheimer’s disease. Neurobiol Aging. 2000;21: concentration with Alzheimer disease progression. JAMA Neurol. 2016;73: 383–421. 60–8. 24. Veerhuis R, Nielsen HM, Tenner AJ. Complement in the brain. Mol Immunol. 52. Tapiola T, et al. Cerebrospinal fluid {beta}-amyloid 42 and tau proteins as 2011;48:1592–603. biomarkers of Alzheimer-type pathologic changes in the brain. Arch Neurol. 25. Finehout EJ, Franck Z, Lee KH. Complement protein isoforms in CSF as 2009;66:382–9. possible biomarkers for neurodegenerative disease. Dis Markers. 2005;21:93– 53. Zwan M, et al. Concordance between cerebrospinal fluid biomarkers and 101. [11C]PIB PET in a memory clinic cohort. J Alzheimers Dis. 2014;41:801–7. 26. Bonham LW, Desikan RS, Yokoyama JS. The relationship between complement factor C3, APOE ε4, amyloid and tau in Alzheimer’s disease. Acta Neuropathol Commun. 2016;4:65. Publisher’s Note 27. Toledo JB, et al. Low levels of cerebrospinal fluid complement 3 and factor Springer Nature remains neutral with regard to jurisdictional claims in H predict faster cognitive decline in mild cognitive impairment. Alzheimers published maps and institutional affiliations. Res Ther. 2014;6:36–10. 28. Smyth MD, Cribbs DH, Tenner AJ, Shankle WR. Decreased levels of C1q in cerebrospinal fluid of living Alzheimer patients correlate with disease state. Neurobiol Aging. 1994;15:609–14. 29. McGeer PL, Walker DG, Pitas RE, Mahley RW, McGeer EG. Apolipoprotein E4 (ApoE4) but not ApoE3 or ApoE2 potentiates β-amyloid protein activation of complement in vitro. Brain Res. 1997;749:135–8. 30. Eikelenboom P, Stam FC. Immunoglobulins and complement factors in senile plaques. Acta Neuropathol. 1982;57:239–42. 31. Hong S, et al. Complement and microglia mediate early synapse loss in Alzheimer mouse models. Science. 2016;352:712–6. 32. Keren-Shaul H, et al. A unique microglia type associated with restricting development of Alzheimer’s disease. Cell. 2017;169:1276–1290.e17. 33. Krasemann S, et al. The TREM2-APOE pathway drives the transcriptional phenotype of dysfunctional microglia in neurodegenerative diseases. Immunity. 2017;47:566–581.e9. 34. Orre M, et al. Isolation of glia from Alzheimer’s mice reveals inflammation and dysfunction. Neurobiol Aging. 2014;35:2746–60. 35. Hoozemans JJM, Veerhuis R, Rozemuller JM, Eikelenboom P. Neuroinflammation and regeneration in the early stages of Alzheimer’s disease pathology. Int J Dev Neurosci. 2006;24:157–65. 36. Cudaback E, Yang Y, Montine TJ, Keene CD. APOE genotype-dependent modulation of astrocyte chemokine CCL3 production. Glia. 2015;63:51–65.