CSF proteome profiling reveals protein panels reflecting the pathophysiological diversity of Alzheimer’s disease

Marta del Campo; Carel FW Peeters; Erik C B Johnson; Lisa Vermunt; Yanaika S Hok-A-Hin; Mirrelijn van Nee; Alice Chen-Plotkin; David J Irwin; William T Hu; James J Lah; Nicholas T Seyfried; Eric B Dammer; Gonzalo Herradon; Lieke H Meeter; John van Swieten; Daniel Alcolea; Alberto Lleó; Allan I Levey; Afina W Lemstra; Yolande A L Pijnenburg; Pieter J Visser; Betty M Tijms; Wiesje M van der Flier; Charlotte E Teunissen

doi:10.1038/s43587-022-00300-1

. Author manuscript; available in PMC: 2023 Nov 10.

Published in final edited form as: Nat Aging. 2022 Nov 10;2(11):1040–1053. doi: 10.1038/s43587-022-00300-1

CSF proteome profiling reveals protein panels reflecting the pathophysiological diversity of Alzheimer’s disease

Marta del Campo ^1,^2,³, Carel FW Peeters ^4,⁵, Erik C B Johnson ⁶, Lisa Vermunt ^1,⁷, Yanaika S Hok-A-Hin ¹, Mirrelijn van Nee ⁴, Alice Chen-Plotkin ⁸, David J Irwin ⁸, William T Hu ^9,¹⁰, James J Lah ⁶, Nicholas T Seyfried ^6,¹⁰, Eric B Dammer ^6,¹⁰, Gonzalo Herradon ², Lieke H Meeter ¹¹, John van Swieten ¹¹, Daniel Alcolea ¹², Alberto Lleó ¹², Allan I Levey ⁶, Afina W Lemstra ⁷, Yolande A L Pijnenburg ⁷, Pieter J Visser ⁷, Betty M Tijms ⁷, Wiesje M van der Flier ^4,⁷, Charlotte E Teunissen ¹

¹Neurochemistry Laboratory, Department of Clinical Chemistry, Amsterdam Neuroscience, Amsterdam University Medical Centers, Location VUmc, Amsterdam, The Netherlands.

²Departamento de Ciencias Farmacéuticas y de la Salud, Facultad de Farmacia, Universidad San Pablo-CEU, CEU Universities, Urbanización Montepríncipe, 28660 Boadilla del Monte, Spain.

³Barcelonaβeta Brain Research Center (BBRC), Pasqual Maragall Foundation, Barcelona, Spain

⁴Department of Epidemiology & Data Science, Amsterdam Public Health research institute, Amsterdam University Medical Centers, Location VUmc, Amsterdam, The Netherlands.

⁵Mathematical & Statistical Methods group (Biometris), Wageningen University & Research, Wageningen, The Netherlands

⁶Goizueta Alzheimer’s Disease Research Center, Emory University, Atlanta, GA, USA.

⁷Alzheimer Center Amsterdam, Department of Neurology, Amsterdam Neuroscience, Amsterdam University Medical Centers, Location VUmc, Amsterdam, The Netherlands.

⁸Department of Neurology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA; Department of Neurology, Emory University School of Medicine, Atlanta, GA, USA

⁹Rutgers-RWJ Medical School, Institute for Health, Health Care Policy, and Aging Research, Rutgers Biomedical and Health Sciences, New Brunswick, NJ, USA

¹⁰Department of Biochemistry, Emory University School of Medicine, Atlanta, GA, USA

¹¹Department of Neurology and Alzheimer Center, Erasmus Medical Center Rotterdam, Rotterdam, Netherlands

¹²Department of Neurology, Institut d'Investigacions Biomèdiques Sant Pau - Hospital de Sant Pau, Universitat Autònoma de Barcelona, Hospital de la Santa Creu i Sant Pau, Barcelona, Catalunya, Spain

^✉

Corresponding author: Marta del Campo Milan, Barcelonaβeta Brain Research Center, Barcelona, Spain, mcampo@barcelonabeta.org

Author Contribution Statement

M.C and C.T conceived of the study. M.C, C.P, L.V and M.vN performed the statistical analysis.

L.V, A. C-P, D.j.I, W.T.H, L. H. M, J.vS, D.A, A.L, A.W. L, Y.A.L.P, P.J.V, B.T.M and W.M.F recruited participants and collected clinical data and samples. M.C and Y.H arranged and prepared samples for proteomics analysis. J.J. L, N.T.S, E.B.D, and A.I.L provided the proteomic data (both Olink and mass spectrometry) from previous published studies13,25. M.C and C.T drafted de manuscript. All authors contributed to revision and editing of the manuscript.

PMCID: PMC10292920 NIHMSID: NIHMS1902789 PMID: 37118088

Abstract

Development of disease-modifying therapies against Alzheimer’s disease (AD) requires biomarkers reflecting the diverse pathological pathways specific for AD. We measured 665 proteins in 797 cerebrospinal fluid (CSF) samples from patients with mild cognitive impairment with abnormal amyloid (MCI(Aβ+): n=50), AD-dementia (n=230), non-AD dementias (n=322) and cognitively unimpaired controls (n=195) using proximity ligation-based immunoassays. Here we identified >100 CSF proteins dysregulated in MCI(Aβ+) or AD compared to controls or non-AD dementias. Proteins dysregulated in MCI(Aβ+) were primarily related to protein catabolism, energy metabolism and oxidative stress, while those specifically dysregulated in AD dementia were related to cell remodeling, vascular function and immune system. Classification modeling unveiled biomarker panels discriminating clinical groups with high accuracies (AUC: 0.85-0.99), which were translated into custom multiplex assays and validated in external and independent cohorts (AUC: 0.8-0.99). Overall, this study provides novel pathophysiological leads delineating the multifactorial nature of AD and potential biomarker tools for diagnostic settings or clinical trials.

Alzheimer’s disease (AD) is the most common age-related neurodegenerative dementia accounting for 60–80% of demented patients and it is an important threat to the aging population. Other common dementia types include dementia with Lewy bodies (DLB) and frontotemporal dementia (FTD), of which the symptomatology and pathology often overlap with that observed in AD patients¹, leading to up to 30% misdiagnosis² and challenging the efficacy of clinical trials³. Several independent pre-clinical and genetic studies indicate that AD is likely caused by a combination of multiple factors and pathways (e.g., immunity, lipid metabolism, vascular dysfunction, endocytic pathway)^4,5 beyond the amyloid cascade, underscoring that AD is a multifactorial disorder. Understanding the biological diversity associated with AD is essential for the development of efficient diagnostic tools and disease-modifying therapies^3,6.

Cerebrospinal fluid (CSF) reflects the ante-mortem biochemical alterations occurring in the brain, and can thus provide the pathobiological fingerprint of different neurodegenerative disorders⁶. The core AD CSF biomarkers used to date reflect the classical hallmarks of AD pathology: Aβ₄₂ or its ratio (Aβ _42/40) reflects senile plaque pathology, while hyperphosphorylated Tau (pTau) and tTau reflect neurofibrillary tangle (NFT) formation and axonal degeneration⁷. These pathological markers are widely considered as good biomarkers for early diagnosis of AD for research purposes and in clinical practice and trials⁸. However, these markers do not capture the biological diversity of AD and have limited utility to demarcate AD from other non-AD dementias, as the latter can present AD type co-pathology^7,9,10. A detailed study of the CSF proteome can unveil novel markers reflecting the multifactorial pathophysiology of AD, improve AD biological definition and open also new insights into potential therapeutic targets and surrogate endpoints for clinical trials targeting different mechanisms^6,11.

Previous mass-spectrometry based studies have provided in depth information of the AD CSF proteome (>1000 detected proteins) using approximately 20-30 samples per group^12,13. Immuno-based technologies (e.g., ELISA, MSD, Simoa) allow higher-throughput screening of samples within a delimited protein array (n < 10 proteins). Novel high-throughput multiplex proteomics platforms (e.g., immune-based proximity extension assay (PEA)) can measure many proteins (>900 proteins) in large data sets with high reproducibility¹⁴. We envision that the use of such immunoassay-based platform allows an initial wide screening with a swift translation of results into custom multiplex immunoassays for subsequent validation in independent cohorts, and thus facilitate bench-to-bedside translation of biomarker findings¹⁵. We here employed such innovative technology to analyse multiple CSF proteins reflecting a wide range of mechanisms in an unprecedented large and well-characterized cohort including patients with mild cognitive impairment with Alzheimer’s pathological changes (MCI(Aβ+)), AD dementia and patients with other major dementias (DLB and FTD) as well as cognitively normal controls without Alzheimer’s pathological changes. Our first aim was to define novel CSF proteomic changes underling AD pathogenesis. Our second aim was to identify and validate biomarkers that could aid in the early and specific biological definition of AD as a complement to the current core biomarkers. We leveraged these findings into custom multiplex assays and validated them in an independent cohort.

Results

An overview of the study design is presented in figure 1. The demographic characteristics and AD CSF biomarker levels are described in table 1. Sex did not differ across groups. Controls were younger in the discovery and validation cohort 2.

Table 1.

Demographic characteristics

	Discovery cohort (proteome profiling)				Validation cohort 1 (proteome profiling)		Validation cohort 2 (Custom PEA assays)
	CON (n=195)	MCI(Aβ+) (n=50)	AD (n=230)	Non-AD dementias¹ (n=322)	CON (n=44)	AD (n=18)	CON (n=110)	MCI(Aβ+) (n=101)	AD (n=88)	Non-AD dementias¹ (n=214)
Age, years (Mean, SD)	58 (8)^a,b,c	67 (7)^d	66 (8)^d	66 (9)^d	68 (7)	65 (12)	60 (6)^a,b,c	70 (5)^d	69 (6)^d	71 (8)^d
Sex (F/M)	79/116	22/28	95/135	116/206	23/21	8/10	54/56	47/54	49/39	69/145
MMSE/MoCA (Mean, SD) ²	28 (2)^a,b,c	26 (2)^b,c	20 (5)^a,c,d	23 (6)^a,b,d	27(2)^b	14(6)^d	29 (1)^a,b,c	26 (3)^b,c,d	21 (4)^a,c,d	23 (5)^a,b,d
APOE4 (+/n, %) ³	44/169 (26%)	33/50 (66%)	133/221 (60%)	102/273 (37%)	5/18 (27%)	12/18 (67%)	29/110 (26%)	61/96 (64%)	47/88 (54%)	67/208 (32%)
CSF Aβ42, pg/mL ⁴	1122 (220)^a,b,c	608 (147)^c,d	608 (123)^c,d	922 (448)^a,b,d	567 (140)^b	218 (116)^d	1283 (301)^a,b/1194(503)^a,b,c	654 (165)^a,b/517(266)^c,d	560 (165)^a,b/537(206)^c,d	980 (463)^a,b/710(549)^a,b,d
CSF tTau, pg/mL ⁴	212 (98)^a,b,c	519 (385)^b,c,d	744 (430)^a,c,d	313 (193)^a,b,d	40 (17)^b	122 (83)^d	295 (250)^a,b/257(125)^a,b,c	572 (310)^a,b/603(408)^c,d	656 (485)^a,b/740(439)^c,d	318 (237)^a,b/359(289)^a,b,d
CSF pTau, pg/mL ⁴	38 (15)^a,b,c	73 (49)^b,c,d	92 (38)^a,c,d	48 (24)^a,b,d	23 (9)^b	60 (23)^d	54 (28)^a,b/38(21)^a,b,c	82 (35)^c,b/96(57)^c,d	89 (44)^c,b/117(84)^c,d	48 (23)^a,b/47(39)^a,b,d

Open in a new tab

Data are median (interquartile range) unless otherwise specified. Between-group analyses were performed using two-sided univariate analysis of variance or Pearson’s chi-square test in normally distributed data with Bonferroni post-hoc adjustment. Analysis of covariance was performed for CSF biomarker analysis adjusting for age and sex when appropiate. Non-Gaussian distributed data were analyzed using Kruskal-Wallis Test. Adjustment for multiple testing was performed using Bonferroni method.

Non-AD dementias included 123 DLB and 199 FTLD patients.

MMSE score was used as a measure of cognitive function in the discovery cohort and validation cohort 2 and it was missing for 39 and 9 participants respectively. MoCA was used as a measure of cognitive function in the validation cohort 1 and was missing for 4 participants.

APOE status was missing for 84 participants in the discovery cohort, 26 and 12 in the validation cohorts 1 and 2 respectively.

⁴

Biomarker data coming from Luminex analysis in the discovery cohort was transformed using appropiate Passing-Babock transformation formulas (see methods). Biomarker data of validation cohort 1 are obtained with Luminex. In validation cohort 2 biomarker values are measured with Innogenetics or Lumipulse for ADC and SPIN respectively. Biomarker values measured with elecsy platform in a subset of cases from ADC (31CON, 2 MCI(A) and 44 non-AD-dem) are not reported in the table.

Between-group analyses of demographic variables were performed using two-sided univariate analysis of variance or Pearson’s chi-square test in normally distributed data with Bonferroni post-hoc adjustment. Non-Gaussian distributed data were analyzed using Kruskal-Wallis Test. Adjustment for multiple testing was performed using Bonferroni method . Adjusted p < 0.05 vs.MCI(Aβ+)^a, vs. AD^b, vs. non-ADdem^c or vs. CON^d.

MCI(Aβ+), mild cognitive impairment with Aβ pathological changes; AD, Alzheimer’s disease; non-Addem, non-AD dementias; SD, Standard desviation; F, Female;M, Male.

CSF proteomic differences in MCI(Aβ+) and AD compared to controls

First, we evaluated which proteins were differentially regulated between either MCI(Aβ+) or AD patients and controls. We identified 112 CSF proteins that were differentially regulated between MCI(Aβ+) and controls (figure 1B). Two proteins were replicates measured within the proteomic platform (see methods) and showed comparable differences (supplementary fig. 1). Most of the proteins were upregulated in MCI(Aβ+) (n = 92, q < 0.05, figure 1B, Source Data Table 1). The top 5 differentially regulated proteins (q < 10⁻⁰⁶) are involved in protein glycosylation (ENTPD5)¹⁶, oxidative stress (PARK7)¹⁷, lysosomal function (PRCP, DPP7)^18,19 and immune system (CRTAM)²⁰. CHIT1 showed the strongest effect (β = 1.1-fold-change: 2; q = 0.004) followed by PARK7, PRCP, IDUA and DPP7 (β > 0.6; fold-change >1.5; figure 1B and Source Data Table 1).

Next, we identified 288 CSF proteins that were differentially regulated between AD dementia patients and controls (figure 1C). Here, 7 proteins were replicates measured within the proteomic platform and showed similar differences (supplementary fig. 1). Most proteins were upregulated in AD (n = 281, q < 0.05, figure 1C, Source Data Table 1). The top 5 differentially regulated proteins (q values < 1⁻¹⁷) are involved in cytoskeletal and extracellular remodeling (ABL1, TMSB10, MMP-10)^21-23, exosome assembly (SDC4)²⁴ and immune system (ITGB2)²³. CHIT1 showed again the strongest effect (β: 0.98, fold-change: 1.9; q = 1,52⁻⁰⁵, Source Data Table 1) followed by TMSB10, CLEC5A, ITGB2, and MMP-10 (β > 0.6, fold-change >1.5; Source Data Table 1). When comparing the results obtained in the discovery cohort with the independent external validation cohort (AD vs control, validation step 1²⁵), we observed concordant results (i.e., consistency in the proteins that are dysregulated and non-dysregulated in both cohorts based on p-values) in 431 (67%) out of the 642 proteins analyzed in both studies (62% after FDR correction, Source Data Table 1, supplementary fig. 2). The strongest protein changes identified in the discovery cohort were also significant in the external validation cohort (Source Data Table 1), and the p-values and fold-changes obtained in the validation cohort correlated positively to those obtained in the discovery cohort (r > 0.65, supplementary fig. 2). Results from the discovery cohort were compared to the CSF proteome information previously reported by two mass-spectrometry (MS)-based studies that analyzed different multidisease¹³ and AD²⁵ CSF cohorts, including the independent external validation cohort described above (validation step 1: CSF cohort 2²⁵). As expected, not all proteins that were dysregulated in our PEA discovery set were detectable by MS approaches (i.e., 99 overlapping proteins with the MS multidisease cohort ¹³, 112 with the MS AD CSF cohort 1²⁵ and 159 with the CSF cohort 2 ²⁵, supplementary table 2). Of note, the same direction of changes (Beta coefficients) was observed in 90% of the proteins analyzed within discovery and CSF cohort 2 when the same PEA platform was used (validation step 1, supplementary fig. 2). However, this % decreased to 54-59% when the proteins dysregulated in the PEA discovery cohort were compared to the same proteins detected by mass-spectrometry in different CSF cohorts^13,25 including the same CSF cohort 2^13,25. These percentages ranged from 57-72% when the same proteins were compared between the different MS studies (supplementary table 2).

CSF proteomic differences between AD and non-AD dementias

To study disease specificity, we next compared the CSF protein expression between AD and non-AD dementias (Of note, 80% of FTD or DLB patients were diagnosed either by autopsy or mutation carriership or had a negative AD CSF biomarker profile). We identified 469 CSF proteins that were differentially regulated between AD and non-AD dementias (figure 1D, Source Data Table 1). Fourteen of these markers were replicates measured within the proteomic platform and showed similar differences (supplementary fig. 1). Most proteins were upregulated in AD and only 16 proteins had lower levels in the CSF of AD patients (q < 0.05, figure 1D, Source Data Table 1). The top 5 differentially regulated proteins (q values < 10⁻¹⁶) are involved in distinct processes, i.e., in cytoskeletal remodeling (ABL1)²¹, energy metabolic processes (ENO2)²⁶, neuropeptide degradation (THOP1)²⁷, oxidative stress (DDAH1)²⁸ and vascular function (RSPO3)²⁸. RSPO3 showed the strongest effect size (β: 0.62, fold-change: 1.5; Source Data Table 1) followed by TMSB10, CRH, LAIR-2, and SMOC2 (β > 0.5, fold-change > 1.4, Source Data Table 1). When results were compared to the CSF MS data from multidisease cohorts ¹³, we again observed that 56% proteins showed the same direction of change between AD and a group with different non-AD neurodegenerative diseases (FTD, Parkinson’s disease and amyotrophic lateral sclerosis, supplementary table 2).

Unique and overlapping proteins between clinical groups

We questioned whether the changes identified in the separate analyses were unique or common across clinical AD stages (MCI(Aβ+) and dementia) and specific for AD. Venn diagram shows that out of the 281 unique proteins that were differentially regulated between AD and controls, the majority were abnormal only in the AD dementia stage (n=189; e.g., ITGB2, SMOC-2, LOX-1, GLO1, BLM hydrolase; figure 2A, B). Of these proteins, 22% (n=63) were also dysregulated in MCI(Aβ+) patients (e.g., ABL1, DDAH1, SDC4, THOP1, PARK7), indicating that a subset of proteins is already dysregulated in early stages of AD (figure 2A, B). We also observed 41 proteins that were dysregulated in MCI(Aβ+) but not in AD dementia patients (e.g., ENTPD5, CRTAM, IDUA, DPP7, PCSK9), suggesting that each disease stage may have a distinctive CSF profile with dysregulated proteins following different trajectories along the disease process (figure 2A, B). We observed that 89% (n=252) of the proteins changed in the comparison of AD with controls were also changed between AD and non-AD dementias. Of these, only one protein (DDC) was more prominently dysregulated in the non-AD group than in the AD group. This data indicates that the far majority of identified proteins are specifically dysregulated in AD (figure 2A, B). A limited number of proteins were dysregulated in both MCI(Aβ+) and AD and had comparable levels to those observed in non-AD dementia patients (n=29; e.g., CCL3, GH, PRDX1, CHIT1, PRCP, KYNU; figure 2A, B), which may represent general biomarkers of neurodegeneration.

Figure 2. — a) Venn diagram depicting the overlap between proteins dysregulated between MCI and controls as well as between AD and controls and non-AD dementias. Markers that were not differentially regulated between MCI or AD and controls are not included. This resulted into four specific protein groups: markers changed specifically in AD (red), markers changed in both MCI(Aβ+) and AD (orange), markers changed only in MCI(Aβ+) but not in AD (yellow), and markers changed in MCI(Aβ+) and AD or AD only, but not between AD and non-AD dementias (grey). The total number of proteins and the name of the top 5 dysregulated proteins in each of the groups is annotated b) Protein trajectories of each specific grouped defined within the Venn diagram. For each individual protein, the log₂ fold change was calculated by subtracting the mean NPX values (protein levels) of the control group from the mean NPX values in each diagnostic group. Lines connect the fold changes of each individual protein along AD disease stage. Dots represent the median and interquartile range of the fold changes for each specific clinical group. Dotted black line represent a 0-fold change. c) Bar graphs depicting the biological pathways enriched in each of the protein groups. Functional enrichment was performed using Metascape selecting GO Biological Processes as ontology source. Terms with a p-value < 0.01, a minimum count of 3, and an enrichment factor > 1.5 were collected and grouped into clusters based on their membership similarities. p-values were calculated based on the accumulative hypergeometric distribution. Kappa scores are used as the similarity metric when performing hierarchical clustering on the enriched terms, and sub-trees with a similarity of > 0.3 are considered a cluster. The most statistically significant term within a cluster is chosen to represent the cluster. The corresponding GO number and biological process is defined in the right side. Stronger colours represent higher significant enrichment. Vertical line represents the significant threshold (unadjusted p <0.01). Stronger colors represent higher significant enrichment. MCI(Aβ+), mild cognitive impairment with Aβ pathological changes; AD, Alzheimer’s disease; non-AD dem, non-AD dementias; CON, cognitively unimpaired controls

CSF proteomic differences reflect distinct biological processes

We next investigated the biological processes reflected by the dysregulated protein subsets along the AD stages (e.g., MCI(Aβ+) and AD dementia, figure 2B, C). Pathways such as protein catabolic processes, proteoglycan metabolism and cell polarity were enriched in the proteins that were dysregulated in MCI(Aβ+) only. Proteins dysregulated in both MCI(Aβ+) and AD dementia were reflecting biological processes related reactive oxygen species (ROS) metabolic processes, protein complex assembly, autophagy, neurotransmitter transport, Wnt signaling pathway, development of peripheral nervous system, energy metabolism, peptidyl serine phosphorylation and nitric oxide biosynthetic processes. Proteins that were dysregulated only at the dementia stage were enriched in biological processes related to immune system and vascular function (e.g., leukocyte cell-adhesion, tumor necrosis factor signaling, establishment endothelial barrier), regulation of Aβ formation, extracellular and cellular morphology (e.g., cell-adhesion mediated by integrins) and neuronal function (e.g., post-synapse organization, neuron guidance). Markers that were differentially regulated in MCI(Aβ+), AD and non-AD dementia, i.e., the potential cross-dementia biomarkers, were enriched in processes related to immune function (e.g., hemopoiesis, viral entry, response to bacterium, immune system processes), chemotaxis and cell killing. The biological processes enriched in each of the protein subsets described here did not appear when a random set of proteins from the proteomic library was selected, supporting the association of the identified processes to AD (supplementary fig. 3).

CSF protein panels discriminate MCI(Aβ+) and AD from controls or non-AD dementias

To translate our findings into practical biomarker tools for routine diagnostics and clinical trials, we next performed classification analysis, followed by internal cross-validation, to identify the minimal number of biomarkers with the maximal discriminative power for AD (CSF panels, figure 1A).

The results revealed a panel of 8 CSF proteins that discriminated AD from controls with very high accuracy (AD-diagnostic panel, AUC 0.96, 95%CI: 0.92-0.99, figure 3A, table 2, supplementary fig. 4). The model contained proteins that were dysregulated in both MCI(Aβ+) and AD dementia stages (ABL1, SDC4, CLEC5A) or only in AD dementia (MMP-10, ITGB2, TREM1). The data-driven model also selected proteins with modest or no significant differences between AD and controls (e.g., THBD, SPON2). All the proteins except THBD were increased in AD compared to non-AD dementias (table 2, supplementary fig. 4). The proteins involved in this panel are related to different pathways including cellular remodeling and protein phosphorylation (MMP10, ABL1), exosome assembly (SDC4), immune system (ITGB2, CLEC5A, TREM1, SPON2) and vascular function (THBD), which were the same pathways as those identified in the differential expression coupled to ontology analysis. We performed correlation analysis in the complete discovery cohort to understand how these markers and the overall panel relate to classical AD CSF biomarkers and cognitive function. All proteins except THBD and SPON2 showed weak to modest correlations with classical AD CSF biomarkers or MMSE scores (figure 3C). ITGB2, ABL1 and MMP-10 were the markers that showed the strongest associations with CSF Aβ₄₂ and tau forms, and MMSE score respectively. Similarly, the biomarker composite score of this panel showed modest correlations with AD CSF biomarkers and MMSE score (supplementary figure 5). APOE4 carrier status did not modify the performance of the model (AUC: 0.96 in APOE4+ and 0.95 in APOE4-, supplementary figure 6). External validation using an independent cohort from Emory University (validation step 1) showed that this model could again discriminate AD patients from controls with high accuracy (AUC 0.94, figure 3A). This AD CSF panel could also discriminate MCI(Aβ+) from controls with good accuracy (AUC 0.87, 95%CI: 0.75-0.97; figure 3D, supplementary figure 7). This panel discriminated AD from non-AD dementias with somewhat lower performance (AD vs non-AD dementias AUC: 0.80, 95%CI: 0.72-0.87; figure 3D, supplementary figure 7).

Figure 3. — Receiver operating characteristic (ROC) curves depicting the performance of CSF biomarker panels discriminating AD (a, orange) from controls in the discovery (CON = 195, AD = 230) and validation 1 (CON=44, AD=18) cohorts. b) ROC curve analyses depicting the performance of CSF biomarker panels discriminating AD (n=230) from non-AD dementias (blue, n=322) in the discovery cohort. Black line is the mean area under the curve (AUC) over all re-samplings (1000 repeats of 5-fold cross-validation, grey lines). Inserts outline corresponding AUC and 95% CI c) Correlation matrix heatmap representing the Spearman’s correlation coefficient in-between the proteins selected in each panel, the classical AD CSF biomarkers and ratios and MMSE score within the complete discovery cohort. d) Overview of all mean AUCs of each panel for the discrimination of the different groups of interest in the discovery CSF proteome profiling (CON=195, MCI(Aβ+) = 50, AD = 230, non-AD dementia = 322) and with the custom assays in validation step 2 (CON=110, MCI(Aβ+) = 101, AD = 88, non-AD dementias = 214). Error bars represent 95% CI. e) Scatter plots show the correlation between the beta-coefficients obtained in the discovery phase to those obtained with the custom assays in validation step 2. Insert indicate the spearman correlation coefficient. f) ROC curves depicting the performance of CSF biomarker panel discriminating AD (n=88) from controls (n=110) or non-AD dementias (n=214) using the custom assays. Inserts outline corresponding AUC and 95% CI. *ABL1 and ITGB2 are proteins that are present in both CSF panels. MCI(Aβ+), mild cognitive impairment with Aβ pathological changes; AD, Alzheimer’s disease; non-AD dem: non-AD dementias; CON, cognitively unimpaired controls.

Table 2.

CSF Markers selected for each of the specific diagnostic panels after 10 fold-cross validation

AD vs. CON panel (n=425)						AD vs. Non-AD dem panel (n=552)
Protein	Effect	fold-change	FDR adjusted p- value	Frecuency 10- fold CV (%)	Overlapped groups	Protein	Effect	fold-change	FDR adjusted p- value	Frecuency 10- fold CV (%)	Overlapped groups
ABL1	0.445	1.361	6.57E-23	100	MCI(Aβ+) and AD	ABL1	0.304	1.235	3.05E-29	100	MCI(Aβ+) and AD
SDC4	0.454	1.370	1.18E-19	100	MCI(Aβ+) and AD	THOP1	0.306	1.236	4.27E-21	100	MCI(Aβ+) and AD
MMP-10	0.616	1.532	1.18E-19	100	AD only	ENO2	0.230	1.173	4.27E-21	100	MCI(Aβ+) and AD
ITGB2	0.639	1.557	5.97E-19	100	AD only	GZMB	−0.109	0.927	9.05E-05	100	n.a.
CLEC5A	0.683	1.605	3.06E-15	100	MCI(Aβ+) and AD	DDC	−0.143	0.906	2.92E-04	100	MCI(Aβ+) and AD
TREM1	0.361	1.285	1.11E-11	100	AD only	MMP7	−0.150	0.901	1.32E-02	80	n.a.
THBD	−0.160	0.895	4.15E-02	100	General dem.	VEGFR-3	−0.013	0.991	5.40E-01	80	n.a.
SPON2	−0.055	0.963	1.90E-01	50	n.a.	ITGB2	0.315	1.244	2.02E-17	70	AD only
						PTK7	−0.015	0.990	5.84E-01	70	MCI(Aβ+) only

Open in a new tab

Table shows the effect (beta coefficient), p- and q- values after applying two-sided nested F-test analysis and FDR post-hoc correction to compare CSF protein abundance between AD and CON or non-AD dementias in the discovery cohort. Only those CSF markers selected within the specific diagnostic panels are included.

The frecuency indicates the fold-based selection proportions for each marker after performing penalized generalized linear modeling (GLM) with an elastic net penalty (see methods).

Overlapped groups indicates whether the proteins were dysregulated in MCI(Aβ+) but not in AD (MCI(Aβ+) only); proteins dysregulated in both MCI(Aβ+) and AD; proteins dysregulated in AD only; or proteins dysregulated in both MCI(Aβ+) and AD or AD only, but not between AD and non AD dementias (general dem) based on the Upset plot (figure 2a). N.a, not applicable as marker was not differentially regulated between AD and controls.

We used the same statistical approach to establish a CSF protein panel demarcating AD patients from patients with non-AD dementias with higher accuracy (AD-differential diagnostic panel). The results showed that a combination of 9 CSF proteins demarcated these dementia types with slightly better performance (AUC 0.87, 95%CI: 0.81-0.93, table 2, figure 3b, supplementary figure 4). As a sensitivity analysis we analyzed the accuracy of this panel in a subset of cases including only patients with genetically or pathologically confirmed non-AD dementia, which resulted in higher performance (AUC 0.96, 95%CI: 0.92-0.996, supplementary fig. 6). Most of the proteins within this panel were highly dysregulated in both the prodromal and AD dementia stages compared to controls and non-AD dementias (e.g., ABL1, THOP1, ENO2, DDC; table 2). Other proteins were dysregulated only in the MCI(Aβ+) (PTK7), or AD dementia stages (ITGB2) compared to controls and non-AD dementias; or only between AD and non AD dementias (GZMB). This panel also included proteins with modest or no significant differences between AD and non-AD dementias (e.g., MMP7, VEGFR-3, PTK7; table 2, supplementary figure 4). It is worth noting that some of the proteins were higher in the non-AD group compared to AD (DDC, GZMB, MMP7). The proteins involved in this panel were related to different pathways including cellular remodeling and protein phosphorylation (MMP7, ABL1, PTK7), neuropeptide degradation (THOP1), energy metabolic processes (ENO2), neurotransmitter synthesis (DDC), the immune system (ITGB2, GZMB) and vascular function (VEGFR-3). Correlation analyses showed that ABL1, ITGB2, THOP1 and ENO2 weakly correlated with CSF Aβ42 (r ≈ −0.2) and showed modest associations with CSF tTau, pTau or the tTau/Aβ42 ratio (r of 0.4 to 0.7; figure 3C). These proteins also significantly, albeit weakly correlated with MMSE score (r ≈ −0.2). The biomarker composite score of this panel showed only weak associations with CSF Tau forms or tTau/Aβ₄₂ ratio and MMSE score (supplementary figure 5). Only two markers (ABL1 and ITGB2) were also present in the AD-diagnostic panel identified previously, suggesting that a different combination of proteins should be used to better demarcate AD from other dementia types. This second panel demarcates MCI(Aβ+) and AD patients from controls with lower performance to that achieved by the AD-diagnostic panel (AUC 0.80 and 0.90 respectively; figure 3D, supplementary figure 7).

Validation of CSF protein panels

To validate the performance of the CSF protein panels, we developed custom multiplex PEA-panels measuring the 15 proteins of the two AD diagnostic panels (table 2). Three protein assays could not be optimally developed due to technical issues (TREM1, GZMB and PTK7). The remaining 12 assays were verified to measure the targeted proteins with optimal coefficients of variation (mean intra- and inter-assay CVs of 5% and 8% respectively) and >90% detectability (supplementary table 3). The validation step 2 performed with an independent multicenter cohort revealed that most of the proteins within the custom panels replicated the changes observed in the high-throughput discovery phase (>60% concordance in the proteins that are dysregulated and non-dysregulated in each group comparison, supplementary table 4). Importantly, we observed that the protein’ fold changes between AD and controls or non-AD dementias strongly correlated with those obtained in the discovery cohort (r = 0.80 and 0.86 respectively, figure 3E), supporting the validity of the findings. The AD-diagnostic panel (now with 7 CSF markers, table 2) showed again high to excellent accuracies, similar to those observed in the discovery phase [AUC_{custom panels} MCI(Aβ+) or AD vs. control: 0.96 and 0.99 respectively; AUC_{custom panels} MCI(Aβ+) or AD vs non-ADdementias: 0.78 and 0.79 respectively, Figure 3D, F]. The accuracy of the AD-differential diagnostic panel (now with 7 CSF markers, table 2) was slightly lower compared to the discovery phase (AUC_{custom panels} AD vs non-AD dementias: 0.80, Figure 3D, F). In summary, translation of the discovery findings into custom assays revealed that the AD-diagnostic panel could be used to discriminate MCI(Aβ+) and AD from controls with accuracies >0.95. This panel could also discriminate AD from non-AD dementias with an accuracy of 0.79.

It is worth noting that when looking into the MCI(Aβ+) vs control, the protein effect’s sizes of the validation step 2 did not strongly correlate with those of the discovery phase (r = 0.37, Figure 3E), suggesting that the differences driven by MCI(Aβ+) group were only partially validated. Indeed, in our discovery phase we also detected a different combination of proteins discriminating MCI(Aβ+) from controls with very high accuracy, (AUC 0.99; 95% CI: 0.97-1, supplementary fig. 8), but this MCI-diagnostic panel was not fully replicated using its corresponding custom assays (AUC 0.85, 95% CI: 0.79-0.9; supplementary fig. 8).

Discussion

This study provides new insights into the multiple and specific protein changes underlying the pathogenesis of AD, and translates the multifactorial findings into practicable CSF biomarker panels. We observed that most of the CSF proteins dysregulated in AD compared to controls were also different when compared to non-AD dementias. The protein changes differed along the AD continuum, reflecting different biological processes at different stages of the disease. We identified, developed and externally validated a 7-CSF protein custom panel discriminating MCI(Aβ+) and AD patients from cognitively unimpaired controls (AUC: 0.96 and 0.99 respectively), and from a group of non-AD dementias (AUC: 0.78 and 0.79 respectively). We also identified and developed a different panel that could differentiate AD from non-AD dementias with higher or similar accuracies depending on the cohort analyzed (AUC: 0.80-0.87). The proteins within the CSF protein panel covered a wide range of mechanisms involved in AD pathogenesis, providing a novel in vivo window to the pathobiological complexity of AD. This study also highlights the effectiveness of our methodological workflow to discover and validate novel biofluid-based biomarkers, leveraging the combination of large well-characterized cohorts with robust and translatable technologies.

Despite the increasing number of CSF proteomics studies performed to date, few have been performed in such a large cohort (665 proteins in 797 samples), and few analyzed the symptomatic spectrum of AD including the MCI stage^25,29,30 . Our study is further unique by including other common dementia types (FTD and DLB) to assess the specificity for AD and advance towards the differential diagnosis of this dementia. Multiple CSF proteins were prominently dysregulated across the different diagnostic groups (> 100 proteins in each paired group comparison; median q-values of 10⁻⁰⁵ or 10⁻¹⁵ among the 20-top proteins dysregulated in the MCI(Aβ+) or AD comparisons respectively). With this approach we confirmed previous results on multiple proteins associated with AD dementia using the same or alternative technologies (e.g., ENO2^12,13,31, CHIT1³², MMP-10^30,32,33, TMSB10²⁹, DDAH1^13,25,29,34, PARK7^12,13,25, SOD1¹² or ITGB2¹³).^13,25 We observed that up to up to 90% of the proteins showed the same direction of changes when comparing the data from the two independent cohorts analyzed with the same PEA technology (i.e., discovery and validation step 1). This % decreased to 54-59% when the PEA discovery results were compared to the same proteins detected by mass-spectrometry in the same CSF cohort used for validation step one (CSF cohort 2) or other MS cohorts^13,25, highlighting that results are to some extent dependent on the platform used. The lack of replication for several dysregulated proteins might be explained by the large differences in sample sizes across studies or the different technologies employed, as unlike MS-approaches detecting relatively small strings of denatured peptides, PEA-technology detects the proteins in their native conformation and can be in principle reach higher analytical sensitivities³⁵. In addition, we observed that not all the proteins dysregulated in our PEA discovery set were detected in previous mass-spectrometry studies, and viceversa^13,25. Therefore, there is generally only partial overlap in proteins that are measured for PEA and MS-platforms and thus, they might be viewed as complementary approaches^35,36.

Most of the proteins dysregulated in AD dementia were also significantly different between AD and non-AD dementias and especially associated to AD. A subset of these markers was also changed in the prodromal stages of the disease (MCI(Aβ+)). Identification of AD-specific changes in early AD stages is important considering the clinicopathological overlap across dementias, resilience factors and the need to develop early and specific treatments for each⁹. Noteworthy, some markers (41) were dysregulated in MCI(Aβ+) but not in AD dementia, suggesting that these are specific for disease stage and that changes may revert with disease progression, a trend that has been already observed for other AD biomarker candidates (MCP-1, VILIP-1, NPTXR)^37-39. These results suggest that CSF proteome follows a highly dynamic trajectory along the disease continuum, which is in line with previous post-mortem brain proteomics studies^25,39,40.

Our study further revealed that the subsets of CSF proteins differentially regulated along the MCI-AD continuum reflect different biological processes. CSF proteins dysregulated only in the early stage of AD (MCI(Aβ+)) were especially related protein catabolism, glycoprotein metabolic processes and cell polarity. These processes were also enriched among the CSF proteins dysregulated in both MCI(Aβ+) and AD patients (e.g., autophagy, generation of metabolites and energy, response to organic cyclic compounds, small molecule catabolic process, or Wnt signaling pathway). Additional processes enriched in both MCI and AD stages included mechanisms related to ROS, protein complex assembly, neurotransmitter transport and protein phosphorylation; all known to play an important role in AD. Alterations in the endolysosomal-autophagic network have been suggested to contribute to disease pathogenesis already in early stages, as observed also by genetic⁴¹ or brain proteomic studies⁴⁰. Several recent brain and CSF mass spectrometry-based proteomics studies similarly show a dysregulation of proteins related to energy metabolism (e.g., glycolysis) and oxidative stress (e.g. ENO2, DDAH, PARK7, SOD1)^{12,13,25,29,31}, some already in the asymptomatic phase of AD^12,13,25. Oxidative stress and glucose metabolism are well known relevant interconnected processes involved in the pathogenesis and progression of AD^5,42, which is in line with our proteome findings. Our results further show that CSF proteins dysregulated only in the AD dementia stage were enriched in a different set of biological processes compared to those observed in early stages such as the immune and vascular systems, extracellular and cellular morphology (including neuronal) and regulation of Aβ formation. Previous brain proteome, genetic and experimental studies support a potential role of these mechanisms in AD pathophysiology^25,40,43-45. Overall, these CSF findings suggest that the changes along AD stages reflect complementary biological processes beyond those observed for the classical AD CSF biomarkers.

To progress towards development of clinical assays, we applied classification analysis and identified two practicable CSF biomarker panels (AD-diagnostic and AD-differential diagnosis) with high predictive performance. The former was validated in an external cohort of AD and non-demented controls with PEA-CSF proteome data (AUCs>0.90). Several proteins within the optimized small panels have been previously identified as potential AD CSF biomarkers (MMP-10^32,33, ENO2^12,13,31 or ITGB2¹³), but several novel candidates were also detected (e.g., ABL1, SDC4, THOP1, SPON2). The relevance of the panels for AD pathology is supported by the association of some specific markers (e.g., ABL1, SDC4 and THOP1) with the classical AD CSF biomarker (r > 0.5 with tTau/Aβ₄₂ ratio). Experimental data indeed showed that proteins such as ABL1 and THOP1 are directly related to Aβ and pTau proteostasis^21,27,46,47. The strongest correlates of cognitive function were MMP-10 and SDC4, which have been previously associated with learning and memory²² or hippocampal synaptic function^24,48. CSF biomarkers within the panel represented different biological processes related to AD (e.g., protein phosphorylation and proteolysis, extracellular and cytoskeletal remodeling, synaptic function, energy metabolism, immune and vascular function) and thus capture the multifactorial nature of AD pathophysiology^4,5,49.

To validate the panels using independent cohorts we have successfully developed custom multiplex assays for 12 out of the 15 selected markers. These custom panels replicated many of the protein changes and effect sizes in the multicenter independent cohort, supporting the robustness of the panels. Importantly, the discriminatory performances of the AD-diagnostic CSF protein panel were also replicated in any of the group comparisons (AUCs>0.80). The AD-differential diagnostic panel showed however lower performance to that obtained within the discovery cohort (AUC_discovery: 0.87, AUC_{customised panel}: 0.80). This panel included markers that were specifically associated with non-AD dementias (e.g., DDC, MMP7) and thus, the heterogeneity of the non-AD dementia diagnoses may influence the results. This, together with the technical failure to measure GZMB and PTK7 may explain the decreased performance of this panel. It is also worth noting that, in contrast to the discovery cohort, only one patient had a confirmed non-AD dementia diagnosis (FTD mutation carriership). Further validation using samples from patients with a confirmed diagnosis of FTD or DLB would be thus needed to establish the optimal performance of this panel, especially considering that the AUC increased to 0.96 when genetically or pathologically confirmed non-AD dementia were analyzed separately in the discovery cohort.

Methodological considerations

This multivariate modeling approach allows to identify the combination of non-redundant biomarkers with the most discriminative and reproducible performance. The higher performance compared to single markers may result from the relationships across the different analytes thereby linking different pathomechanisms involved in AD^15,50,51. This data-driven modeling approach differs from classical approaches in which the top dysregulated biomarkers identified are individually validated in subsequent replication steps. Classical approaches require the identification of markers with high discrimination potential and low overlap between groups. However, the magnitude of change for the majority of the proteins in most of the CSF dementia proteomics studies performed to date, including this one, are small to modest (−0.2 < β < 0.6)^{12,13,25,32,38}, limiting their clinical utility as single markers. We show that with CSF biomarker panels it is possible to reach sufficient diagnostic accuracy. The identification of CSF panels requires, however, a large sample size as well as the use of robust high throughput, reproduceable and translatable technologies, attributes that have been fulfilled within the current study. We hypothesize that the partial validation of our MCI(Aβ+) data with the custom assays might be explained by the heterogeneity of the phenotype of the group. Some of the markers within the panel may thus not be reproducible across cohorts, thereby reducing the corresponding classification performance (from 0.99 to 0.85). Thus, identifying the combination of markers that define complex and heterogeneous phenotypes across different data sets likely requires analysis of large sample sizes. Yet, the MCI(Aβ+) sample size is higher than most proteomics studies performed to date²⁹ and our discovery results aligned with previous findings (e.g., for PARK7, ENO2, DDAH1, SOD1)^13,25. Importantly, the use of an immuno-based technology allowed us to effectively translate the CSF panels detected in our CSF proteome profiling into customized assays for independent validation, thereby overcoming the cross-technology gap often encountered in biomarker studies⁵². An additional advantage of antibody-based platforms that, like immunoassays, the analytical approach is standardized and does not require extensive sample pre-processing, which may ultimately reduce potential preanalytical variability and increase reproducibility across studies. The workflow employed in this study could thus be also relevant for the effective development of biofluid biomarkers within and outside the field of dementias.

Despite the large number of proteins (n=645) and samples (n=797) analyzed, this study is not without limitations. First, some of the CSF proteins measured and known to be changed in AD (e.g., NPTXR³⁸ or CHI3L1⁵³) were not dysregulated in this study. This indicates that our findings are to some extent dependent on the technology employed. Second, we cannot exclude that some MCI cases may progress to a non-AD dementia overtime. However, the MCI group had an intermediate or high likelihood to progress to AD dementia based on classical AD CSF biomarkers levels. Third, we did not include the preclinical and prodromal stages of different dementia types. Fourth, 20-30% of the patients clinically diagnosed with a non-AD dementia had a positive AD CSF biomarker profile and thus, potential misdiagnosis cannot be excluded. Still, we observed that the AD-differential diagnostic panel discriminated AD from the subset of patients with confirmed non-AD dementia with high accuracy (AUC 0.96). Last, the potential added value of the panels to supplement the diagnosis based on the classical AD CSF biomarkers could not be compared in this study, as our clinical group selection was based also on the AD CSF biomarker profiles to minimize potential misdiagnosis. Further studies are needed to understand the potential added value of these panels to classical AD CSF and the very recently developed blood-based biomarkers⁵⁴. Moreover, it would be of interest to validate the novel biomarkers in plasma using this strategy or ultra-sensitive technologies in future studies. Of note, these customized biomarker panels reflect different biological processes associated to AD pathophysiology^4,5,49. This unique quality may make them suitable to improve the biological definition but also to monitor treatment response in the upcoming clinical trials targeting different AD mechanisms beyond amyloid and Tau^6,11.

In summary, this study provides new insights into the CSF proteins and related biological processes that are dysregulated along the symptomatic phases of AD and are specific for this condition, providing new potential leads for etiological and mechanistic studies. In addition, we also identified and validated CSF panels that accurately discriminate AD from controls, or from non-AD dementias. The proteins of these panels are involved in multiple mechanisms associated with AD pathophysiology, therefore reflecting the complexity and multifactorial biology of AD. Importantly, the technology employed allowed us to efficiently translate these panels into customized assays and validate them in an independent cohort, underpinning the effectiveness of the workflow employed for biomarker development. With these custom panels is possible to start to define the potential added value of these markers in routine diagnosis and clinical trials of drugs targeting different pathomechanisms of AD.

Methods

This study was conducted in accordance with Standards for Reporting of Diagnostic Accuracy (STARD) guidelines. The STARD checklist can be found as part of the Supplementary material.

Ethics statement

Ethical approval was given by the Institutional Ethical Review Boards of each center (Discovery cohort: VUmc: AD CSF biobank METC number 00-211; University of Pennsilvania: language and cognitive impairment in parkinson’s disease and parkinson’s disease with dementia or dementia with lewy bodies IRB069801; Emory University: IRB00042049; Erasmus MC: METC numbers 124.378/1993/17 and 202.927/2001/143 or under the Parelsnoer initiative 2009-170. Validation cohort: Emory IRB 00078273 and 00024959, SPIN cohort: COLLECTION 16/2013). All participants gave written informed consent

Participants

The total cohort (n=797) of this prospective study included CSF samples from 50 patients with MCI(Aβ+), 230 with AD, 322 with non-AD dementias (DLB=123 and FTD=199) and 195 cognitively normal controls (CON; table 1). The majority of the samples were selected from the Amsterdam Dementia Cohort (ADC; all groups)⁵⁵. To enrich for dementia cases with confirmed non-AD diagnosis, additional CSF samples from the Center for Neurodegenerative Disease Research at the University of Pennsylvania⁵⁶ (95 non-AD dementias and 20 AD), Erasmus Medical Center (18 non-AD dementia and 1 CON) and the Goizueta Alzheimer’s Disease Research Center at Emory University (4 non-AD dementia) were included. For validation step 1, we employed CSF proteomic results generated with the same immune-based technology at Emory University as part of an independent study (Damer et al, under review)^25,36. This validation cohort included 44 cognitively unimpaired controls and 18 AD dementia cases, and it was used to further support the validity of the protein changes identified and the accuracy of the diagnostic panel discriminating these specific groups. For validation step 2, i.e., validation of the customized panels, an independent multicenter cohort (n=513) contained samples from the ADC (n=260, all groups and different from the discovery cohort)⁵⁵ and the Sant Pau Initiative⁵⁷.

All participants of every cohort underwent standard neurological and cognitive assessments and diagnosis was assigned according to international consensus criteria for MCI⁵⁸, AD⁵⁹, DLB^60,61, and FTD^62-64. Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment test (MoCA) were used as a measure of global cognition. Levels of CSF Aβ₄₂, tTau and pTau(181) (‘AD CSF biomarkers’) were used to support AD diagnoses or AD pathological changes in MCI. These markers were analyzed locally as part of the diagnostic procedure using commercially available kits (VUmc/ErasmusMC: ELISA INNOTEST Aβ(1-42), hTAUAg, phospho-Tau(181P); Fujirebio, Ghent, Belgium; Penn and Emory: Luminex xMAP INNO-BIA AlzBio3; Luminex Corp, Austin, TX; SPIN: Lumipulse G600, Fujirebio)^65-67. Positive CSF AD biomarker profile was defined locally as increased t-tau/Aβ₄₂ in the cohorts from ADC/ErasmusMC (>0.46), Pennsylvania (>0.30); low Aβ₄₂/t-tau in the Emory university cohort (<6); or low Aβ_42/40 ratio (<0.062) and high total tau (>456pg/ml) or p-tau (>63pg/ml) in the SPIN cohort^65-67. All MCI cases were Aβ+ which was defined locally as low Aβ₄₂ in the ADC cohort (<813 pg/ml) or low Aβ_42/40 ratio in SPIN cohort (<0,062)^67,68. In the discovery cohort diagnosis of AD patients was supported by positive AD CSF biomarker profile in 225 patients (98%), 22 and for 6 patients this was additionally confirmed by autopsy. Non-AD dementia was confirmed in 101 patients (31%) either by autopsy (13 DLB and 46 FTD) or the presence of an autosomal dominantly inherited mutation (42 FTD mutation carriers). 157 of the non-AD patients (49%) were clinically diagnosed with DLB or FTD and had a negative CSF AD biomarker profile, 60 patients (19%) were clinically diagnosed with DLB or FTD but had a positive CSF profile and 4 patients (1%) were clinically diagnosed with DLB or FTD and did not have AD CSF biomarkers available. Of note, 21% of the patients with confirmed non-AD dementia had also a positive AD CSF profile. The control group included individuals with subjective cognitive decline, in whom objective cognitive and laboratory investigations were normal (i.e., criteria for MCI, dementia, or any other neurological or psychiatric disorder not fulfilled) with additionally negative AD CSF biomarkers in 193 cases (99%))^55,69.

In both validation cohorts, step 1 and step 2, AD and control cases were supported by a positive or negative AD CSF profile respectively. Some AD cases within the second validation cohort had a negative AD CSF profile or negative Tau pathology based on CSF pTau, and thus were not included in the statistical analysis (n=22). Non-AD dementia patients within the second validation cohort were mostly clinically diagnosed (i.e., only one FTD mutation carrier and none with autopsy confirmation) and 27% had a positive AD CSF biomarker profile. Patient demographics and clinical and biochemical values from all cohorts used in this study are listed in table 1.

CSF protein profiling.

CSF proteins (979) were quantified using the 11 specific and validated multiplex antibody-based protein panels based on the proximity extension assay (PEA) that were available at the time in which the analysis was performed (Cardiometabolic, Cardiovascular II and III, cell regulation, development, immune response, inflammation, metabolism, neurology, oncology II and organ damage; Olink Proteomics, Uppsala, Sweden; supplementary table 1)⁵⁸. Each panel contains reagents to measure up to 92 unique proteins, though 30 proteins can be measured in different panels (replicates). Briefly, for each protein, a unique pair of oligonucleotide-labeled antibody probes bind to the targeted protein, and if the two probes are in close proximity a PCR target sequence is formed by a proximity-dependent DNA polymerization event, which is further quantified using the Fluidigm BioMark^™ HD real-time PCR platform⁵⁸, thereby maximizing sensitivity and specificity. Protein levels are reported in log2-scale as Normalized Protein eXpression (NPX). All characteristics and validation data for each assay are available at the manufacture’s webpage (www.olink.com). Biochemical analyses were performed by the company, and thus researchers performing the experiment were blinded to any sample characteristic or clinical data. Samples were randomized across plates containing appropriate intra- and inter-plate quality controls (QC) from manufacturer and measured in two different rounds. Each round included 16 bridging samples covering different clinical groups which were used for reference sample normalization to control for potential batch effects. Each assay has an experimentally determined lower limit of detection (LOD) estimated as three standard deviations above noise level from the negative controls that are included on every plate. Only proteins with values over the lower limit of detection (LOD) in at least 85% of the samples were selected for further statistical analysis, in which remaining raw values under LOD (2.4% of all measurements) were kept as provided by manufacturer. A total of 665 proteins (642 unique proteins) were ultimately included for statistical analysis of the discovery cohort (supplementary table 2).

Development of custom PEA assays.

Custom designed multiplex PEA assays are developed by the manufacturer following standardized protocols¹⁵. We aimed to develop assays able to measure the 15 proteins selected upon the classification analysis described in section 3.4. TREM1 and PTK7 assays could not be developed due to technical failure. Each plate measured besides the samples of interest, four CSF QC samples, a negative control and three calibrators used for normalization. QC samples and calibrators were measured in triplicate. Each assay has an experimentally determined LOD defined as three standard deviations above noise level. Precision (intra- and inter-assay CV) were calculated using the 4 CSF QC samples. No cross-reactivity between assays was detected. Assay parameters including LOD, detectability and CVs of the 13 assays ultimately developed are included in supplementary table 3. Samples from the step 2 validation cohort were randomized across plates and normalized for any plate effects using the built-in inter-plate controls according to manufacturers’ recommendations. Protein abundance was reported in NPX values. GZMB had missingness (i.e., values < LOD) of 83% was thus not included in the analysis.

Statistical analysis & reproducibility

All processing and analyses were conducted using R version 3.5.3 and SPSS version 25. Between-group analyses of demographic variables were performed using two-sided univariate analysis of variance or Pearson’s chi-square test in normally distributed data. Analysis of covariance was performed for classical AD CSF biomarker analysis adjusting for age and sex when needed. Non-Gaussian distributed data were analyzed using Kruskal-Wallis Test. Adjustment for multiple testing was performed using Bonferroni method. For the CSF proteome data, differences in protein abundance between pairs of clinical groups were evaluated by using nested linear models as previously described⁷⁰, in which for each individual protein feature, we assessed if its addition to a base model containing age and gender contributed to model fit⁷⁰. This concurs with evaluating if, conditional on the effects of age and sex, protein abundance differs between two groups of interest. This approach entails a nested F-test equivalent to the two-sided regression tests. For each pairwise comparison, multiplicity was taken into account by controlling the False Discovery Rate (FDR)⁷¹ at q ≤ 0.05 based on the number of features analyzed. The results obtained in the discovery cohort were compared to i) those obtained with the external cohort of validation step 1 (AD vs. control) and ii) those from the multidisease CSF cohort previously analyzed by mass-spectrometry approaches^13,25 (AD vs. controls (n=35); AD vs. a group of non-AD neurodegenerative disorders (n=60) including FTD, amyotrophic lateral sclerosis -ALS- and Parkinson’s disease -PD-). We analyzed the concordance and overlap (i.e., consistency regarding the proteins that are dysregulated and non-dysregulated between the groups of interest) as well as the correlation of q-values and beta values between discovery and validation cohorts.

We next evaluated which CSF protein combination (CSF panels) could best discriminate the groups of interest while keeping the number of markers to the minimum, so that they can be ultimately translated into small, practical custom panel. For this purpose, binary classification signatures (AD vs. CON, AD vs. non-AD dementias) were constructed by way of penalized generalized linear modeling (GLM) with an elastic net penalty (a linear combination of lasso and ridge penalties)⁷⁰ in the discovery CSF cohort using the glmnet package and including age and sex as covariates⁷². This penalty enables estimation in settings where the feature to sample ratio is too high for standard generalized linear regression. Moreover, it performs automatic feature decorrelation as well as feature-selection. For each classification exercise, we compare multiple models which reflect (a) a grid of values for the elastic-net mixing parameter, reflecting strong decorrelation to a pure logistic lasso regression and (b) a grid of values reflecting the maximum number of proteins that may be selected under each model (21 markers maximum). The former grid (a) takes into account that we have little information on the collinearity burden in the data. The latter grid (b) takes into account that we want to keep the number of selected proteins relatively low for the future development of customized panels. The optimal penalty parameters in the penalized models were determined on the basis of (balanced) 10-fold cross-validation of the model likelihood⁷⁰. The cross-validation was performed with balanced folds, by which each fold has an outcome group ratio close to the corresponding ratio in the full data set, also referred to as stratified cross-validation. Predictive performance of all models was assessed by way of (the comparison of) Receiver Operating Characteristic (ROC) curves and Area Under the ROC Curves (AUCs). The model with the highest AUC and lowest number of markers for each classification signature was selected. The fold-based selection proportions for each marker were assessed to identify and select the most promising markers within each model (i.e., features that are stably selected across each individual fold thereby minimizing potential overfitting). To reflect the manual selection pressure for these final marker sets, each final logistic signature was subjected to a ridge-regularization with a penalty parameter of 0.1. The performance (AUC) was evaluated by internal validation: repeated 5-fold cross-validation with 1000 repeats. The 95% confidence interval around the resulting AUCs was based on resampling quantiles (percentile method). External validations assessed the performance of the final models (as fitted on the discovery CSF cohort) in the validation cohorts.

Partial non-parametric correlation analysis was performed to understand the association among proteins within the CSF panels with the classical AD CSF biomarkers and cognitive function (MMSE score) using the complete discovery cohort without stratifying per diagnostic category conditioning on age and sex as covariates. To further understand how the biomarker information contained in each individual panel relates to classical AD CSF biomarkers and cognitive function, a biomarker composite score was calculated for each panel using principal component analysis (PCA). PCA was performed on the markers of each panel and the corresponding groups for which the specific panel was built. The first principal component was used as the biomarker panel composite score. This composite score was subsequently correlated (spearman correlations on complete observations) to classical AD CSF biomarkers and MMSE score in their corresponding clinical groups adjusting for age and sex. Innotest values generated for the Amsterdam Dementia Cohort were adjusted for drift over time as previously described⁶⁸. UPENN values had lower means for Aβ₄₂ on the Innotest, which were first linearly transformed to normalize to the same mean. Passing-Bablock transformation formulas were calculated based on individuals with both Luminex and Innotest values for Aβ₄₂ (n=32), tTau (n=32) and pTau (n=27) and used the formulas to estimate the equivalent Innotest values for those samples measured with Luminex platform only (transformed_ Aβ₄₂ = (Luminex_ Aβ₄₂*4.65) - 36.23; transformed_tTau = (Luminex_tTau*5.28) - 2.03; transformed_ptau = (Luminex_pTau*1.88) + 27.36).

Functional enrichment analysis was performed using Metascape⁷¹ selecting GO Biological Processes as ontology source. All the CSF proteins optimally analyzed with Olink arrays (n=644 protein gene products) were included as the enrichment background. Default parameters were used for the analysis in which terms with a p-value < 0.01, a minimum count of 3, and an enrichment factor > 1.5 were collected and grouped into clusters based on their membership similarities. No statistical methods were used to pre-determine sample size but our sample size is similar or even higher than those reported in previous publications ^25,29,30

Supplementary Material

Supplement

NIHMS1902789-supplement-Supplement.pptx^{(17.2MB, pptx)}

SourceDataTable 1

NIHMS1902789-supplement-SourceDataTable_1.xlsx^{(117.6KB, xlsx)}

STARD Checklist

NIHMS1902789-supplement-STARD_Checklist.pdf^{(804.3KB, pdf)}

SupplementaryTables1-4

NIHMS1902789-supplement-SupplementaryTables1-4.pdf^{(777.7KB, pdf)}

Acknowledgements

This research is part of the neurodegeneration research program of Amsterdam Neuroscience. This study was supported by: Alzheimer Nederland (WE.03-2018-05, MC and CT) and Selfridges Group Foundation (NR170065, MC and CT). MC and GH are supported by the attraction talent fellowship of Comunidad de Madrid (2018-T2/BMD-11885) and San Pablo CEU University. DA acknowledges support from Institute of Health Carlos III (PI18/00435, INT19/00016) and the Department of Health Generalitat de Catalunya PERIS program (SLT006/17/125). Collection of patients samples and data from Penn University (AC and DI) was supported by different funding sources: National Institute on Aging: NINDS R01-NS109260-01A1, P01-AG066597, P30-AG072979 (formerly P30-AG10124), U19-AG062418-03 (formerly NINDSP50-NS053488-09). Alzheimer Center Amsterdam is supported by Stichting Alzheimer Nederland and Stichting VUmc fonds. The chair WvF is supported by the Pasman stichting. The clinical database structure was developed with funding from Stichting Dioraphte. Research programs of WvF have been funded by ZonMW, NWO, EU-FP7, EU-JPND, Alzheimer Nederland, Hersenstichting CardioVascular Onderzoek Nederland, Health~Holland, Topsector Life Sciences & Health, stichting Dioraphte, Gieskes-Strijbis fonds, stichting Equilibrio, Edwin Bouw fonds, Pasman stichting, stichting Alzheimer & Neuropsychiatrie Foundation, Philips, Biogen MA Inc, Novartis-NL, Life-MI, AVID, Roche BV, Fujifilm, Combinostics. WF holds the Pasman chair. WF is recipient of ABOARD, which is a public-private partnership receiving funding from ZonMW (#73305095007) and Health~Holland, Topsector Life Sciences & Health (PPP-allowance; #LSHM20106). All funding is paid to her institution. Research of CET is supported by the European Commission (Marie Curie International Training Network, grant agreement No 860197 (MIRIADE), Innovative Medicines Initiatives 3TR (Horizon 2020, grant no 831434) EPND ( IMI 2 Joint Undertaking (JU) under grant agreement No. 101034344grant no) and JPND (bPRIDE), National MS Society (Progressive MS alliance) and Health Holland, the Dutch Research Council (ZonMW), Alzheimer Drug Discovery Foundation, The Selfridges Group Foundation, Alzheimer Netherlands, Alzheimer Association. CT is recipient of ABOARD, which is a public-private partnership receiving funding from ZonMW (#73305095007) and Health~Holland, Topsector Life Sciences & Health (PPP-allowance; #LSHM20106). ABOARD also receives funding from Edwin Bouw Fonds and Gieskes-Strijbisfonds. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

Footnotes

Competing Interests Statement

MC has been an invited speaker at Eisai. LV received a grant for CORAL consortium by Olink. BMT and PJV are inventors on a patent (#WO2020197399A1; owned by Stichting VUmc). D.I. is a Scientific Advisory Board Member for Denali Therapeutics. D.A. participated in advisory boards from Fujirebio-Europe and Roche Diagnostics and received speaker honoraria from Fujirebio-Europe, Roche Diagnostics, Nutricia, Krka Farmacéutica S.L., Zambon S.A.U. and Esteve Pharmaceuticals S.A. D.A. declares a filed patent application (WO2019175379 A1 Markers of synaptopathy in neurodegenerative disease). WF has performed contract research for Biogen MA Inc, and Boehringer Ingelheim. WF has been an invited speaker at Boehringer Ingelheim, Biogen MA Inc, Danone, Eisai, WebMD Neurology (Medscape), Springer Healthcare. WF is consultant to Oxford Health Policy Forum CIC, Roche, and Biogen MA Inc. WF participated in advisory boards of Biogen MA Inc and Roche. All funding is paid to her institution. WF is member of the steering committee of PAVE, and Think Brain Health. WF was associate editor of Alzheimer, Research & Therapy in 2020/2021. WF is associate editor at Brain. CET has a collaboration contract with ADx Neurosciences, Quanterix and Eli Lilly, performed contract research or received grants from AC-Immune, Axon Neurosciences, Bioconnect, Bioorchestra, Brainstorm Therapeutics, Celgene, EIP Pharma, Eisai, Grifols, Novo Nordisk, PeopleBio, Roche, Toyama, Vivoryon. She serves on editorial boards of Medidact Neurologie/Springer, Alzheimer Research and Therapy, Neurology: Neuroimmunology & Neuroinflammation, and is editor of a Neuromethods book Springer. She had speaker contracts for Roche, Grifols, Novo Nordisk. The rest of the authors declare no competing interest.

Data availability.

The data that support the findings of this study are available from the authors on reasonable request. Mass spectrometry CSF results from previous studies have been provided by co-authors (J.J.L, N.T.S, E.B.D, and A.I.L) and raw data is available in appropriate repositories ^13,25.

Code availability.

The code that supports the findings of this study is available from the authors on reasonable request. All models were built using publicly available packages and functions in R.

References

1.Jellinger KA Neuropathology of Dementia Disorders. J Alzheimers Dis Parkinsonism Jellinger J Alzheimers Dis Parkinsonism 4, (2014). [Google Scholar]
2.Beach TG, Monsell SE, Phillips LE & Kukull W Accuracy of the clinical diagnosis of Alzheimer disease at National Institute on Aging Alzheimer Disease Centers, 2005-2010. J Neuropathol Exp Neurol 71, 266–273 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Gauthier S. et al. Why has therapy development for dementia failed in the last two decades? Alzheimer’s & Dementia 12, 60–64 (2016). [DOI] [PubMed] [Google Scholar]
4.Jansen IE et al. Genome-wide meta-analysis identifies new loci and functional pathways influencing Alzheimer’s disease risk. Nat Genet 12, 43 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
5.van der Kant R, Goldstein LSB & Ossenkoppele R Amyloid-β-independent regulators of tau pathology in Alzheimer disease. Nat Rev Neurosci (2020) doi: 10.1038/s41583-019-0240-3. [DOI] [PubMed] [Google Scholar]
6.Park SA, Han SM & Kim CE New fluid biomarkers tracking non-amyloid-β and non-tau pathology in Alzheimer’s disease. Experimental and Molecular Medicine vol. 52 556–568 Preprint at 10.1038/s12276-020-0418-9 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Schoonenboom NSM et al. Cerebrospinal fluid markers for differential dementia diagnosis in a large memory clinic cohort. Neurology 78, 47–54 (2012). [DOI] [PubMed] [Google Scholar]
8.Jack CR et al. NIA-AA Research Framework: Toward a biological definition of Alzheimer’s disease. Alzheimer’s & Dementia 14, 535–562 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Robinson JL et al. Neurodegenerative disease concomitant proteinopathies are prevalent, age-related and APOE4-associated. Brain 141, 2181–2193 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Ewers M. et al. CSF biomarkers for the differential diagnosis of Alzheimer’s disease: A large-scale international multicenter study. Alzheimer’s & Dementia 11, 1306–1315 (2015). [DOI] [PubMed] [Google Scholar]
11.van Bokhoven P. et al. The Alzheimer’s disease drug development landscape. Alzheimers Res Ther 13, (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Bader JM et al. Proteome profiling in cerebrospinal fluid reveals novel biomarkers of Alzheimer ‘ s disease. Mol Syst Biol 1–17 (2020) doi: 10.15252/msb.20199356. [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Higginbotham L. et al. Integrated Proteomics Reveals Brain-Based Cerebrospinal Fluid Biomarkers in Asymptomatic and Symptomatic Alzheimer’s Disease. Sci Adv 806752 (2020) doi: 10.1126/sciadv.aaz9360. [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Assarsson E. et al. Homogenous 96-Plex PEA Immunoassay Exhibiting High Sensitivity, Specificity, and Excellent Scalability. PLoS One 9, e95192 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Enroth S. et al. High throughput proteomics identifies a high-accuracy 11 plasma protein biomarker signature for ovarian cancer. Commun Biol 2, (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Fang M. et al. The ER UDPase ENTPD5 promotes protein N-glycosylation, the Warburg effect, and proliferation in the PTEN pathway. Cell 143, 711–724 (2010). [DOI] [PubMed] [Google Scholar]
17.Clements CM, Mcnally RS, Conti BJ, Mak TW & Ting JP protein , stabilizes the antioxidant transcriptional master regulator Nrf2. Proc Natl Acad Sci U S A 103, 15091–6 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Maes MB et al. Kinetic investigation of human dipeptidyl peptidase II (DPPII)-mediated hydrolysis of dipeptide derivatives and its identification as quiescent cell proline dipeptidase (QPP)/dipeptidyl peptidase 7 (DPP7). Biochem J 386, 315–324 (2005). [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Soisson SM et al. Structural definition and substrate specificity of the S28 protease family: The crystal structure of human prolylcarboxypeptidase. BMC Struct Biol 10, 1–8 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Patiño-Lopez G. et al. Human class-I restricted T cell associated molecule is highly expressed in the cerebellum and is a marker for activated NKT and CD8+ T lymphocytes. J Neuroimmunol 171, 145–155 (2006). [DOI] [PubMed] [Google Scholar]
21.Derkinderen P. et al. Tyrosine 394 is phosphorylated in Alzheimer’s paired helical filament tau and in fetal tau with c-Abl as the candidate tyrosine kinase. Journal of Neuroscience 25, 6584–6593 (2005). [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Beroun A. et al. MMPs in learning and memory and neuropsychiatric disorders. Cellular and Molecular Life Sciences vol. 76 3207–3228 Preprint at 10.1007/s00018-019-03180-8 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Erickson-Viitanen S, Ruggieri S, Natalini P & Horecker BL Thymosin beta 10, a new analog of thymosin beta 4 in mammalian tissues. Arch Biochem Biophys 225, 407–413 (1983). [DOI] [PubMed] [Google Scholar]
24.Baietti MF et al. Syndecan-syntenin-ALIX regulates the biogenesis of exosomes. Nat Cell Biol 14, 677–685 (2012). [DOI] [PubMed] [Google Scholar]
25.B Johnson EC et al. Large-scale proteomic analysis of Alzheimer’s disease brain and cerebrospinal fluid reveals early changes in energy metabolism associated with microglia and astrocyte activation. Nat Med (2020) doi: 10.1038/s41591-020-0815-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Isgrò MA, Bottoni P & Scatena R Neuron-Specific Enolase as a Biomarker: Biochemical and Clinical Aspects. Adv Exp Med Biol 867, 125–143 (2015). [DOI] [PubMed] [Google Scholar]
27.Pollio G. et al. Increased expression of the oligopeptidase THOP1 is a neuroprotective response to Aβ toxicity. Neurobiol Dis (2008) doi: 10.1016/j.nbd.2008.04.004. [DOI] [PubMed] [Google Scholar]
28.Dayoub H. et al. Dimethylarginine dimethylaminohydrolase regulates nitric oxide synthesis: genetic and physiological evidence. Circulation 108, 3042–3047 (2003). [DOI] [PubMed] [Google Scholar]
29.Wesenhagen KEJ, Teunissen CE, Visser PJ & Tijms BM Cerebrospinal fluid proteomics and biological heterogeneity in Alzheimer’s disease: A literature review. Critical Reviews in Clinical Laboratory Sciences vol. 57 86–98 Preprint at 10.1080/10408363.2019.1670613 (2019). [DOI] [PubMed] [Google Scholar]
30.Boström G. et al. Different Inflammatory Signatures in Alzheimer’s Disease and Frontotemporal Dementia Cerebrospinal Fluid. Journal of Alzheimer’s Disease 81, 629–640 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Konijnenberg E. et al. APOE ϵ4 genotype-dependent cerebrospinal fluid proteomic signatures in Alzheimer’s disease. Alzheimers Res Ther 12, (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Whelan CD et al. Multiplex proteomics identifies novel CSF and plasma biomarkers of early Alzheimer’s disease. Acta Neuropathol Commun (2019) doi: 10.1186/s40478-019-0795-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Duits FH et al. Matrix Metalloproteinases in Alzheimer’s Disease and Concurrent Cerebral Microbleeds. Journal of Alzheimer’s Disease 48, 711–720 (2015). [DOI] [PubMed] [Google Scholar]
34.Dayon L. et al. Alzheimer disease pathology and the cerebrospinal fluid proteome. Alzheimers Res Ther 10, (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Suhre K, McCarthy MI & Schwenk JM Genetics meets proteomics: perspectives for large population-based studies. Nat Rev Genet 22, 19–37 (2021). [DOI] [PubMed] [Google Scholar]
36.Dammer EB et al. Multi-Platform Proteomic Analysis of Alzheimer’s Disease Cerebrospinal Fluid and Plasma Reveals Network Biomarkers Associated with Proteostasis and the Matrisome. bioRxiv 2022.06.20.494087 (2022) doi: 10.1101/2022.06.20.494087. [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Wildsmith KR et al. Identification of longitudinally dynamic biomarkers in Alzheimer’s disease cerebrospinal fluid by targeted proteomics. Mol Neurodegener 9, 22 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Begcevic I. et al. Brain-related proteins as potential CSF biomarkers of Alzheimer’s disease: A targeted mass spectrometry approach. J Proteomics 182, 12–20 (2018). [DOI] [PubMed] [Google Scholar]
39.Bai B. et al. Deep Multilayer Brain Proteomics Identifies Molecular Networks in Alzheimer’s Disease Progression. Neuron 105, 975–991.e7 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Seyfried NT et al. A Multi-network Approach Identifies Protein-Specific Co-expression in Asymptomatic and Symptomatic Alzheimer’s Disease. Cell Syst (2017) doi: 10.1016/j.cels.2016.11.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Gao S, Casey AE, Sargeant TJ & Mäkinen V-P Genetic variation within endolysosomal system is associated with late-onset Alzheimer’s disease Song. Brain 141, 2711–2720 (2018). [DOI] [PubMed] [Google Scholar]
42.Butterfield DA & Halliwell B Oxidative stress, dysfunctional glucose metabolism and Alzheimer disease. Nature Reviews Neuroscience vol. 20 148–160 Preprint at 10.1038/s41583-019-0132-6 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Butkiewicz M. et al. Genetic meta-analysis of diagnosed Alzheimer’s disease identifies new risk loci and implicates Aβ, tau, immunity and lipid processing. Nat Genet (2019) doi: 10.1038/s41588-019-0358-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Lepelletier FX, Mann DMA, Robinson AC, Pinteaux E & Boutin H Early changes in extracellular matrix in Alzheimer’s disease. Neuropathol Appl Neurobiol 43, 167–182 (2017). [DOI] [PubMed] [Google Scholar]
45.Wang H. et al. Regulation of beta-amyloid production in neurons by astrocyte-derived cholesterol. Proc Natl Acad Sci U S A 118, (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Philibert KD, Marr RA, Norstrom EM & Glucksman MJ Identification and characterization of Aβ peptide interactors in Alzheimer’s disease by structural approaches. Front Aging Neurosci 6, 1–9 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Lonskaya I, Hebron ML, Desforges NM, Franjie A & Moussa CE-H. Tyrosine kinase inhibition increases functional parkin- B eclin-1 interaction and enhances amyloid clearance and cognitive performance. EMBO Mol Med 5, 1247–1262 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Siddiqui TJ et al. An LRRTM4-HSPG Complex Mediates Excitatory Synapse Development on Dentate Gyrus Granule Cells. (2013) doi: 10.1016/j.neuron.2013.06.029. [DOI] [PubMed] [Google Scholar]
49.Gong C-X, Liu F & Iqbal K Perspective Multifactorial Hypothesis and Multi-Targets for Alzheimer’s Disease. Journal of Alzheimer’s Disease 64, 107–117 (2018). [DOI] [PubMed] [Google Scholar]
50.Johnstone D, Milward EA, Berretta R & Moscato P Multivariate Protein Signatures of Pre-Clinical Alzheimer’s Disease in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) Plasma Proteome Dataset. PLoS One 7, e34341 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
51.Kim Y. et al. Targeted proteomics identifies liquid-biopsy signatures for extracapsular prostate cancer. Nat Commun 7, (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Teunissen CE, Otto M, Engelborghs S, Herukka S-K & Lehmann S White paper by the Society for CSF Analysis and Clinical Neurochemistry: Overcoming barriers in biomarker development and clinical translation. Alzheimer’s Research & Therapy (2018) 12, 23 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Teunissen CE et al. Novel diagnostic cerebrospinal fluid protein biomarkers for pathologic subtypes of frontotemporal dementia identified by proteomics. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring 9, (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Teunissen CE et al. Blood-based biomarkers for Alzheimer’s disease: towards clinical implementation. Lancet Neurol (2021) doi: 10.1016/S1474-4422(21)00361-6. [DOI] [PubMed] [Google Scholar]
55.van der Flier WM & Scheltens P Amsterdam Dementia Cohort: Performing Research to Optimize Care. Journal of Alzheimer’s Disease 62, 1091–1111 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Lleó A. et al. A 2-Step Cerebrospinal Algorithm for the Selection of Frontotemporal Lobar Degeneration Subtypes. JAMA Neurol (2018) doi: 10.1001/jamaneurol.2018.0118. [DOI] [PMC free article] [PubMed] [Google Scholar]
57.Del Campo M. et al. Recommendations to standardize preanalytical confounding factors in Alzheimer’s and Parkinson’s disease cerebrospinal fluid biomarkers: an update. Biomark Med 6, 419–30 (2012). [DOI] [PubMed] [Google Scholar]
58.Albert MS et al. The diagnosis of mild cognitive impairment due to Alzheimer’s disease: Recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement 7, 270–279 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
59.Dubois B. et al. Research criteria for the diagnosis of Alzheimer’s disease: revising the NINCDS-ADRDA criteria. Lancet neurology 6, 734–46 (2007). [DOI] [PubMed] [Google Scholar]
60.McKeith IG et al. Diagnosis and management of dementia with Lewy bodies: third report of the DLB Consortium. Neurology 65, 1863–72 (2005). [DOI] [PubMed] [Google Scholar]
61.McKeith IG et al. Diagnosis and management of dementia with Lewy bodies: Fourth consensus report of the DLB Consortium. Neurology 89, 88–100 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Rascovsky K. et al. Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain 134, 2456–77 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
63.Armstrong MJ et al. Criteria for the diagnosis of corticobasal degeneration. Neurology 80, 496–503 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Litvan I. et al. Clinical research criteria for the diagnosis of progressive supranuclear palsy (Steele-Richardson-Olszewski syndrome): report of the NINDS-SPSP international workshop. Neurology 47, 1–9 (1996). [DOI] [PubMed] [Google Scholar]
65.Shaw LM et al. Cerebrospinal fluid biomarker signature in Alzheimer’s disease neuroimaging initiative subjects. Ann Neurol 65, 403–413 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Duits FH et al. The cerebrospinal fluid ‘alzheimer profile’: Easily said, but what does it mean? Alzheimer’s and Dementia 10, 713–723.e2 (2014). [DOI] [PubMed] [Google Scholar]
67.Alcolea D. et al. The Sant Pau Initiative on Neurodegeneration (SPIN) cohort: A data set for biomarker discovery and validation in neurodegenerative disorders. Alzheimer’s and Dementia: Translational Research and Clinical Interventions 5, 597–609 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Tijms BM et al. Unbiased approach to counteract upward drift in cerebrospinal fluid amyloid-β 1–42 analysis results. Clin Chem 64, 576–585 (2018). [DOI] [PubMed] [Google Scholar]
69.Jessen F. et al. A conceptual framework for research on subjective cognitive decline in preclinical Alzheimer’s disease. Alzheimers Dement 10, 844–52 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Leeuw FA et al. Blood-based metabolic signatures in Alzheimer’s disease. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring 8, 196–207 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
71.Benjamini Y & Hochberg Y Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society B vol. 57 289–300 Preprint at 10.2307/2346101 (1995). [DOI] [Google Scholar]
72.Friedman J, Hastie T & Tibshirani R Regularization paths for generalized linear models via coordinate descent. J Stat Softw 33, 1–22 (2010). [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement

NIHMS1902789-supplement-Supplement.pptx^{(17.2MB, pptx)}

SourceDataTable 1

NIHMS1902789-supplement-SourceDataTable_1.xlsx^{(117.6KB, xlsx)}

STARD Checklist

NIHMS1902789-supplement-STARD_Checklist.pdf^{(804.3KB, pdf)}

SupplementaryTables1-4

NIHMS1902789-supplement-SupplementaryTables1-4.pdf^{(777.7KB, pdf)}

Data Availability Statement

The code that supports the findings of this study is available from the authors on reasonable request. All models were built using publicly available packages and functions in R.

[R1] 1.Jellinger KA Neuropathology of Dementia Disorders. J Alzheimers Dis Parkinsonism Jellinger J Alzheimers Dis Parkinsonism 4, (2014). [Google Scholar]

[R2] 2.Beach TG, Monsell SE, Phillips LE & Kukull W Accuracy of the clinical diagnosis of Alzheimer disease at National Institute on Aging Alzheimer Disease Centers, 2005-2010. J Neuropathol Exp Neurol 71, 266–273 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R3] 3.Gauthier S. et al. Why has therapy development for dementia failed in the last two decades? Alzheimer’s & Dementia 12, 60–64 (2016). [DOI] [PubMed] [Google Scholar]

[R4] 4.Jansen IE et al. Genome-wide meta-analysis identifies new loci and functional pathways influencing Alzheimer’s disease risk. Nat Genet 12, 43 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R5] 5.van der Kant R, Goldstein LSB & Ossenkoppele R Amyloid-β-independent regulators of tau pathology in Alzheimer disease. Nat Rev Neurosci (2020) doi: 10.1038/s41583-019-0240-3. [DOI] [PubMed] [Google Scholar]

[R6] 6.Park SA, Han SM & Kim CE New fluid biomarkers tracking non-amyloid-β and non-tau pathology in Alzheimer’s disease. Experimental and Molecular Medicine vol. 52 556–568 Preprint at 10.1038/s12276-020-0418-9 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R7] 7.Schoonenboom NSM et al. Cerebrospinal fluid markers for differential dementia diagnosis in a large memory clinic cohort. Neurology 78, 47–54 (2012). [DOI] [PubMed] [Google Scholar]

[R8] 8.Jack CR et al. NIA-AA Research Framework: Toward a biological definition of Alzheimer’s disease. Alzheimer’s & Dementia 14, 535–562 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R9] 9.Robinson JL et al. Neurodegenerative disease concomitant proteinopathies are prevalent, age-related and APOE4-associated. Brain 141, 2181–2193 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R10] 10.Ewers M. et al. CSF biomarkers for the differential diagnosis of Alzheimer’s disease: A large-scale international multicenter study. Alzheimer’s & Dementia 11, 1306–1315 (2015). [DOI] [PubMed] [Google Scholar]

[R11] 11.van Bokhoven P. et al. The Alzheimer’s disease drug development landscape. Alzheimers Res Ther 13, (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R12] 12.Bader JM et al. Proteome profiling in cerebrospinal fluid reveals novel biomarkers of Alzheimer ‘ s disease. Mol Syst Biol 1–17 (2020) doi: 10.15252/msb.20199356. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R13] 13.Higginbotham L. et al. Integrated Proteomics Reveals Brain-Based Cerebrospinal Fluid Biomarkers in Asymptomatic and Symptomatic Alzheimer’s Disease. Sci Adv 806752 (2020) doi: 10.1126/sciadv.aaz9360. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R14] 14.Assarsson E. et al. Homogenous 96-Plex PEA Immunoassay Exhibiting High Sensitivity, Specificity, and Excellent Scalability. PLoS One 9, e95192 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R15] 15.Enroth S. et al. High throughput proteomics identifies a high-accuracy 11 plasma protein biomarker signature for ovarian cancer. Commun Biol 2, (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R16] 16.Fang M. et al. The ER UDPase ENTPD5 promotes protein N-glycosylation, the Warburg effect, and proliferation in the PTEN pathway. Cell 143, 711–724 (2010). [DOI] [PubMed] [Google Scholar]

[R17] 17.Clements CM, Mcnally RS, Conti BJ, Mak TW & Ting JP protein , stabilizes the antioxidant transcriptional master regulator Nrf2. Proc Natl Acad Sci U S A 103, 15091–6 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R18] 18.Maes MB et al. Kinetic investigation of human dipeptidyl peptidase II (DPPII)-mediated hydrolysis of dipeptide derivatives and its identification as quiescent cell proline dipeptidase (QPP)/dipeptidyl peptidase 7 (DPP7). Biochem J 386, 315–324 (2005). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R19] 19.Soisson SM et al. Structural definition and substrate specificity of the S28 protease family: The crystal structure of human prolylcarboxypeptidase. BMC Struct Biol 10, 1–8 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R20] 20.Patiño-Lopez G. et al. Human class-I restricted T cell associated molecule is highly expressed in the cerebellum and is a marker for activated NKT and CD8+ T lymphocytes. J Neuroimmunol 171, 145–155 (2006). [DOI] [PubMed] [Google Scholar]

[R21] 21.Derkinderen P. et al. Tyrosine 394 is phosphorylated in Alzheimer’s paired helical filament tau and in fetal tau with c-Abl as the candidate tyrosine kinase. Journal of Neuroscience 25, 6584–6593 (2005). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R22] 22.Beroun A. et al. MMPs in learning and memory and neuropsychiatric disorders. Cellular and Molecular Life Sciences vol. 76 3207–3228 Preprint at 10.1007/s00018-019-03180-8 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R23] 23.Erickson-Viitanen S, Ruggieri S, Natalini P & Horecker BL Thymosin beta 10, a new analog of thymosin beta 4 in mammalian tissues. Arch Biochem Biophys 225, 407–413 (1983). [DOI] [PubMed] [Google Scholar]

[R24] 24.Baietti MF et al. Syndecan-syntenin-ALIX regulates the biogenesis of exosomes. Nat Cell Biol 14, 677–685 (2012). [DOI] [PubMed] [Google Scholar]

[R25] 25.B Johnson EC et al. Large-scale proteomic analysis of Alzheimer’s disease brain and cerebrospinal fluid reveals early changes in energy metabolism associated with microglia and astrocyte activation. Nat Med (2020) doi: 10.1038/s41591-020-0815-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R26] 26.Isgrò MA, Bottoni P & Scatena R Neuron-Specific Enolase as a Biomarker: Biochemical and Clinical Aspects. Adv Exp Med Biol 867, 125–143 (2015). [DOI] [PubMed] [Google Scholar]

[R27] 27.Pollio G. et al. Increased expression of the oligopeptidase THOP1 is a neuroprotective response to Aβ toxicity. Neurobiol Dis (2008) doi: 10.1016/j.nbd.2008.04.004. [DOI] [PubMed] [Google Scholar]

[R28] 28.Dayoub H. et al. Dimethylarginine dimethylaminohydrolase regulates nitric oxide synthesis: genetic and physiological evidence. Circulation 108, 3042–3047 (2003). [DOI] [PubMed] [Google Scholar]

[R29] 29.Wesenhagen KEJ, Teunissen CE, Visser PJ & Tijms BM Cerebrospinal fluid proteomics and biological heterogeneity in Alzheimer’s disease: A literature review. Critical Reviews in Clinical Laboratory Sciences vol. 57 86–98 Preprint at 10.1080/10408363.2019.1670613 (2019). [DOI] [PubMed] [Google Scholar]

[R30] 30.Boström G. et al. Different Inflammatory Signatures in Alzheimer’s Disease and Frontotemporal Dementia Cerebrospinal Fluid. Journal of Alzheimer’s Disease 81, 629–640 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R31] 31.Konijnenberg E. et al. APOE ϵ4 genotype-dependent cerebrospinal fluid proteomic signatures in Alzheimer’s disease. Alzheimers Res Ther 12, (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R32] 32.Whelan CD et al. Multiplex proteomics identifies novel CSF and plasma biomarkers of early Alzheimer’s disease. Acta Neuropathol Commun (2019) doi: 10.1186/s40478-019-0795-2. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R33] 33.Duits FH et al. Matrix Metalloproteinases in Alzheimer’s Disease and Concurrent Cerebral Microbleeds. Journal of Alzheimer’s Disease 48, 711–720 (2015). [DOI] [PubMed] [Google Scholar]

[R34] 34.Dayon L. et al. Alzheimer disease pathology and the cerebrospinal fluid proteome. Alzheimers Res Ther 10, (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R35] 35.Suhre K, McCarthy MI & Schwenk JM Genetics meets proteomics: perspectives for large population-based studies. Nat Rev Genet 22, 19–37 (2021). [DOI] [PubMed] [Google Scholar]

[R36] 36.Dammer EB et al. Multi-Platform Proteomic Analysis of Alzheimer’s Disease Cerebrospinal Fluid and Plasma Reveals Network Biomarkers Associated with Proteostasis and the Matrisome. bioRxiv 2022.06.20.494087 (2022) doi: 10.1101/2022.06.20.494087. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R37] 37.Wildsmith KR et al. Identification of longitudinally dynamic biomarkers in Alzheimer’s disease cerebrospinal fluid by targeted proteomics. Mol Neurodegener 9, 22 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R38] 38.Begcevic I. et al. Brain-related proteins as potential CSF biomarkers of Alzheimer’s disease: A targeted mass spectrometry approach. J Proteomics 182, 12–20 (2018). [DOI] [PubMed] [Google Scholar]

[R39] 39.Bai B. et al. Deep Multilayer Brain Proteomics Identifies Molecular Networks in Alzheimer’s Disease Progression. Neuron 105, 975–991.e7 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R40] 40.Seyfried NT et al. A Multi-network Approach Identifies Protein-Specific Co-expression in Asymptomatic and Symptomatic Alzheimer’s Disease. Cell Syst (2017) doi: 10.1016/j.cels.2016.11.006. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R41] 41.Gao S, Casey AE, Sargeant TJ & Mäkinen V-P Genetic variation within endolysosomal system is associated with late-onset Alzheimer’s disease Song. Brain 141, 2711–2720 (2018). [DOI] [PubMed] [Google Scholar]

[R42] 42.Butterfield DA & Halliwell B Oxidative stress, dysfunctional glucose metabolism and Alzheimer disease. Nature Reviews Neuroscience vol. 20 148–160 Preprint at 10.1038/s41583-019-0132-6 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R43] 43.Butkiewicz M. et al. Genetic meta-analysis of diagnosed Alzheimer’s disease identifies new risk loci and implicates Aβ, tau, immunity and lipid processing. Nat Genet (2019) doi: 10.1038/s41588-019-0358-2. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R44] 44.Lepelletier FX, Mann DMA, Robinson AC, Pinteaux E & Boutin H Early changes in extracellular matrix in Alzheimer’s disease. Neuropathol Appl Neurobiol 43, 167–182 (2017). [DOI] [PubMed] [Google Scholar]

[R45] 45.Wang H. et al. Regulation of beta-amyloid production in neurons by astrocyte-derived cholesterol. Proc Natl Acad Sci U S A 118, (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R46] 46.Philibert KD, Marr RA, Norstrom EM & Glucksman MJ Identification and characterization of Aβ peptide interactors in Alzheimer’s disease by structural approaches. Front Aging Neurosci 6, 1–9 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R47] 47.Lonskaya I, Hebron ML, Desforges NM, Franjie A & Moussa CE-H. Tyrosine kinase inhibition increases functional parkin- B eclin-1 interaction and enhances amyloid clearance and cognitive performance. EMBO Mol Med 5, 1247–1262 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R48] 48.Siddiqui TJ et al. An LRRTM4-HSPG Complex Mediates Excitatory Synapse Development on Dentate Gyrus Granule Cells. (2013) doi: 10.1016/j.neuron.2013.06.029. [DOI] [PubMed] [Google Scholar]

[R49] 49.Gong C-X, Liu F & Iqbal K Perspective Multifactorial Hypothesis and Multi-Targets for Alzheimer’s Disease. Journal of Alzheimer’s Disease 64, 107–117 (2018). [DOI] [PubMed] [Google Scholar]

[R50] 50.Johnstone D, Milward EA, Berretta R & Moscato P Multivariate Protein Signatures of Pre-Clinical Alzheimer’s Disease in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) Plasma Proteome Dataset. PLoS One 7, e34341 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R51] 51.Kim Y. et al. Targeted proteomics identifies liquid-biopsy signatures for extracapsular prostate cancer. Nat Commun 7, (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R52] 52.Teunissen CE, Otto M, Engelborghs S, Herukka S-K & Lehmann S White paper by the Society for CSF Analysis and Clinical Neurochemistry: Overcoming barriers in biomarker development and clinical translation. Alzheimer’s Research & Therapy (2018) 12, 23 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R53] 53.Teunissen CE et al. Novel diagnostic cerebrospinal fluid protein biomarkers for pathologic subtypes of frontotemporal dementia identified by proteomics. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring 9, (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R54] 54.Teunissen CE et al. Blood-based biomarkers for Alzheimer’s disease: towards clinical implementation. Lancet Neurol (2021) doi: 10.1016/S1474-4422(21)00361-6. [DOI] [PubMed] [Google Scholar]

[R55] 55.van der Flier WM & Scheltens P Amsterdam Dementia Cohort: Performing Research to Optimize Care. Journal of Alzheimer’s Disease 62, 1091–1111 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R56] 56.Lleó A. et al. A 2-Step Cerebrospinal Algorithm for the Selection of Frontotemporal Lobar Degeneration Subtypes. JAMA Neurol (2018) doi: 10.1001/jamaneurol.2018.0118. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R57] 57.Del Campo M. et al. Recommendations to standardize preanalytical confounding factors in Alzheimer’s and Parkinson’s disease cerebrospinal fluid biomarkers: an update. Biomark Med 6, 419–30 (2012). [DOI] [PubMed] [Google Scholar]

[R58] 58.Albert MS et al. The diagnosis of mild cognitive impairment due to Alzheimer’s disease: Recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement 7, 270–279 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R59] 59.Dubois B. et al. Research criteria for the diagnosis of Alzheimer’s disease: revising the NINCDS-ADRDA criteria. Lancet neurology 6, 734–46 (2007). [DOI] [PubMed] [Google Scholar]

[R60] 60.McKeith IG et al. Diagnosis and management of dementia with Lewy bodies: third report of the DLB Consortium. Neurology 65, 1863–72 (2005). [DOI] [PubMed] [Google Scholar]

[R61] 61.McKeith IG et al. Diagnosis and management of dementia with Lewy bodies: Fourth consensus report of the DLB Consortium. Neurology 89, 88–100 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R62] 62.Rascovsky K. et al. Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain 134, 2456–77 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R63] 63.Armstrong MJ et al. Criteria for the diagnosis of corticobasal degeneration. Neurology 80, 496–503 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R64] 64.Litvan I. et al. Clinical research criteria for the diagnosis of progressive supranuclear palsy (Steele-Richardson-Olszewski syndrome): report of the NINDS-SPSP international workshop. Neurology 47, 1–9 (1996). [DOI] [PubMed] [Google Scholar]

[R65] 65.Shaw LM et al. Cerebrospinal fluid biomarker signature in Alzheimer’s disease neuroimaging initiative subjects. Ann Neurol 65, 403–413 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R66] 66.Duits FH et al. The cerebrospinal fluid ‘alzheimer profile’: Easily said, but what does it mean? Alzheimer’s and Dementia 10, 713–723.e2 (2014). [DOI] [PubMed] [Google Scholar]

[R67] 67.Alcolea D. et al. The Sant Pau Initiative on Neurodegeneration (SPIN) cohort: A data set for biomarker discovery and validation in neurodegenerative disorders. Alzheimer’s and Dementia: Translational Research and Clinical Interventions 5, 597–609 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R68] 68.Tijms BM et al. Unbiased approach to counteract upward drift in cerebrospinal fluid amyloid-β 1–42 analysis results. Clin Chem 64, 576–585 (2018). [DOI] [PubMed] [Google Scholar]

[R69] 69.Jessen F. et al. A conceptual framework for research on subjective cognitive decline in preclinical Alzheimer’s disease. Alzheimers Dement 10, 844–52 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R70] 70.Leeuw FA et al. Blood-based metabolic signatures in Alzheimer’s disease. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring 8, 196–207 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R71] 71.Benjamini Y & Hochberg Y Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society B vol. 57 289–300 Preprint at 10.2307/2346101 (1995). [DOI] [Google Scholar]

[R72] 72.Friedman J, Hastie T & Tibshirani R Regularization paths for generalized linear models via coordinate descent. J Stat Softw 33, 1–22 (2010). [PMC free article] [PubMed] [Google Scholar]

PERMALINK

CSF proteome profiling reveals protein panels reflecting the pathophysiological diversity of Alzheimer’s disease

Marta del Campo

Carel FW Peeters

Erik C B Johnson

Lisa Vermunt

Yanaika S Hok-A-Hin

Mirrelijn van Nee

Alice Chen-Plotkin

David J Irwin

William T Hu

James J Lah

Nicholas T Seyfried

Eric B Dammer

Gonzalo Herradon

Lieke H Meeter

John van Swieten

Daniel Alcolea

Alberto Lleó

Allan I Levey

Afina W Lemstra

Yolande A L Pijnenburg

Pieter J Visser

Betty M Tijms

Wiesje M van der Flier

Charlotte E Teunissen

Abstract

Results

Figure 1. Study overview and differential abundance of CSF proteins across groups.

Table 1.

CSF proteomic differences in MCI(Aβ+) and AD compared to controls

CSF proteomic differences between AD and non-AD dementias

Unique and overlapping proteins between clinical groups

Figure 2. Differentially regulated CSF proteins change along AD spectrum and reflect different biological processes.

CSF proteomic differences reflect distinct biological processes

CSF protein panels discriminate MCI(Aβ+) and AD from controls or non-AD dementias

Figure 3. CSF biomarker panels for early and specific diagnosis of AD.

Table 2.

Validation of CSF protein panels

Discussion

Methodological considerations

Methods

Ethics statement

Participants

CSF protein profiling.

Development of custom PEA assays.

Statistical analysis & reproducibility

Supplementary Material

Acknowledgements

Footnotes

Data availability.

Code availability.

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases