Comparison of Commonly Measured Plasma and Cerebrospinal Fluid Proteins and Their Significance for the Characterization of Cognitive Impairment Status

Abstract

Background:

Although cerebrospinal fluid (CSF) amyloid-β₄₂ peptide (Aβ₄₂) and phosphorylated tau (p-tau) and blood p-tau are valuable for differential diagnosis of Alzheimer’s disease (AD) from cognitively normal (CN) there is a lack of validated biomarkers for mild cognitive impairment (MCI).

Objective:

This study sought to determine how plasma and CSF protein markers compared in the characterization of MCI and AD status.

Methods:

This cohort study included Alzheimer’s Disease Neuroimaging Initiative (ADNI) participants who had baseline levels of 75 proteins measured commonly in plasma and CSF (257 total, 46 CN, 143 MCI, and 68 AD). Logistic regression, least absolute shrinkage and selection operator (LASSO) and Random Forest (RF) methods were used to identify the protein candidates for the disease classification.

Results:

We observed that six plasma proteins panel (APOE, AMBP, C3, IL16, IGFBP2, APOD) outperformed the seven CSF proteins panel (VEGFA, HGF, PRL, FABP3, FGF4, CD40, RETN) as well as AD markers (CSF p-tau and Aβ₄₂) to distinguish the MCI from AD [area under the curve (AUC) = 0.75 (plasma proteins), AUC = 0.60 (CSF proteins) and AUC = 0.56 (CSF p-tau and Aβ₄₂)]. Also, these six plasma proteins performed better than the CSF proteins and were in line with CSF p-tau and Aβ₄₂ in differentiating CN versus MCI subjects [AUC = 0.89 (plasma proteins), AUC = 0.85 (CSF proteins) and AUC = 0.89 (CSF p-tau and Aβ₄₂)]. These results were adjusted for age, sex, education, and APOE ϵ4 genotype.

Conclusions:

This study suggests that the combination of 6 plasma proteins can serve as an effective marker for differentiating MCI from AD and CN.

Keywords

Alzheimer’s disease biomarkers mild cognitive impairment proteins

INTRODUCTION

Alzheimer’s disease (AD) is one of the most common type of dementia, and still lacks therapies that are definitive in lowering its morbidity [1, 2]. The discovery of AD biomarkers for early detection that is vital for successful disease regulation. For clinical trials, they can accelerate accurate screening and lower the burden of large, lengthy, and costly trials, and also overcome uncertainty in patient progression on treatment outcome measures [3, 4]. The two major pathogenic components of AD are plaques, composed of amyloid-β (Aβ), and neurofibrillary tangles, composed of hyperphosphorylated tau [5, 6]. Several metabolic processes like inflammation, oxidative damage, and lysosomal dysfunction interact with the aberrant protein deposition [2, 7]. The three biomarkers in cerebrospinal fluid (CSF) consistently studied and validated are Aβ₄₂, phosphorylated tau (p-tau), and total tau protein (t-tau). These biomarkers are good predictors for differentiating AD from cognitively normal (CN) patients with a mean sensitivity of 82% and specificity of 82% for Aβ₄₂, sensitivity of 80% and specificity of 90% for t-tau, and sensitivity of 80% and specificity of 83% for p-tau [8]. Combination of CSF biomarkers with structural or functional brain imaging markers may provide higher diagnostic accuracy than the CSF biomarkers or imaging biomarkers alone [9]. However, CSF biomarkers Aβ₄₂, p-tau, and t-tau are inadequate in distinguishing mild cognitive impairment (MCI) from CN subjects and AD from MCI subjects. Additionally, lumbar puncture to collect CSF is less accepted by the older adults, especially MCI patients [10]. Although there is an emerging pipeline of drugs for the treatment of MCI and early-stage AD, and the US Food and Drug Administration (FDA) has approved the anti-amyloid-beta protofibril antibody lecanemab (https://doi.org/10.1002/trc2.12295), the challenge remains that lack of effective biomarkers to diagnose and monitor MCI.

The National Institute on Aging and Alzheimer’s Association (NIA-AA) research framework for AD in 2018 defined AD biologically, by neuropathologic change or biomarkers, and treated cognitive impairment as a symptom/sign of the disease rather than the definition of the disease. This theory states that rather than relying solely on in vivo clinical diagnostic criteria, additional confirmatory evidence is required from positron emission tomography (PET) or fluid biomarkers of AD specific Aβ and tau pathology [8 , 11–13]. Patients with AD, however, also have damage to their blood–brain barrier (BBB) which will led the identification of blood-based biomarkers possible, through the measurement of proteins that have passed through the disrupted BBB [14]. Recent studies suggest that the plasma Aβ₄₂, and p-tau biomarkers accurately predict the change in cognition and subsequent AD dementia in cognitively unimpaired patients [15 –17]. Plasma p-tau₂₁₇ alone in BioFinder cohort was reported to accurately predict (area under the curve (AUC) = 0.83) AD progression within 4 years in those with subjective cognitive decline and MCI subjects [16]. The AUC was improved if p-tau₂₁₇ combined with the test score of memory, executive function and APOE ϵ4 genotype [16, 18].

The above scientific evidence suggests that the ATN (amyloid/tau/neurodegeneration) biomarkers in CSF and blood have demonstrated usefulness in AD. Researchers has also identified other protein markers in blood distinct form Aβ and tau that also performed well in AD classification and prediction [19 –23]. Improvement in MCI diagnosis may benefit from considering these other biomarkers under the umbrella of ATN biomarkers or in addition to them. As the population ages and more older adults develop AD dementia, we need to develop tests that are cost effective and can be widely used to aid accurate diagnosis of MCI and AD. In the present study, we aimed to compare commonly measured protein markers in plasma and CSF to make differential diagnoses of cognitive status at baseline and compared them with established AD biomarkers namely CSF Aβ₄₂ and p-tau.

MATERIAL AND METHODS

Data source

Data used in this study were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) phase 1 (ADNI1) database (https://adni.loni.usc.edu/). ADNI was launched in 2003 as a public-private partnership led by principal investigator Michael W. Weiner, MD. The overall objective of ADNI was to develop and validate in vivo biomarkers for AD. Detailed methodology and description of ADNI can been found on the ADNI website. The 63 ADNI sites in the USA and Canada received approval from their local Institutional Review Boards, and each participant at each site provided written informed consent. The ADNI study consist of 4 phases: ADNI1, ADNIGO, ADNI2, and ADNI3 [24]. Plasma and CSF proteomics data, however, is only available in the ADNI1. Therefore, the ADNI1 database was the only source of participants for this study (Fig. 1).

Fig. 1

Participant selection flow chart and study design.

Participants

We included subjects (n = 257) with plasma and CSF proteomics data from the ADNI1 database. In ADNI, diagnoses of CN, MCI, and AD dementia are based on established research diagnostic criteria [25, 26]. The general cognition was assessed by Mini-Mental State Examination (MMSE) [27, 28]. Clinical Dementia Rating (CDR) Sum of Boxes (CDRSB) was also used to assess overall severity of clinical impairment [29, 30]. We also obtained demographic and clinical data, including age, sex, education years, APOE ϵ4 genotype, CSF p-tau and Aβ₄₂ from “adnimerge” Table in ADNIMERGE R package [28 , 32].

Measurements of the plasma and CSF proteins

We obtained the plasma and CSF proteomics data from ADNI Biomarkers Consortium project for plasma and CSF QC Multiplex data sheets, which were downloaded (zip files) from the ADNI website (https://ida.loni.usc.edu/pages/access/studyData.jsp?categoryId=11&subCategoryId=33). These samples were interrogated on Luminex xMAP platform by Rules-Based Medicine (RBM), to measure the levels of 190 analytes in plasma and 159 analytes in CSF using a multiplex immunoassay panel. The Luminex xMAP technology used a flow-based laser apparatus to detect fluorescent polystyrene microspheres which are loaded with different ratios of two spectrally distinct fluorochromes. Using a precise ratio of the fluorochromes, up to 100 different beads can be generated such that each contains a unique color-coded signature. The beads serve as a solid phase matrix that can then be coated with either ligand or capture antibodies and then standard sandwich or competitive assay formats applied to detect the analytes. Samples were obtained in the morning following an overnight fast at the baseline visit in the ADNI1 study. For the majority of samples, the time from collection to freezing was within 120 (plasma) and 60 (CSF) min. Processing, aliquoting, and storage at 80°C were performed according to the ADNI Biomarker Core Laboratory Standard Operating Procedures. The quality check (QC) procedure was performed by ADNI on all the plasma and CSF analytes. Assays detail and quantification methods were provided in the data primer given in zip files described above. We observed that there were 75 commonly measured proteins in plasma and CSF. So, we obtained a subset of 75 proteins (Table 1) from plasma and CSF data for further analyses and compare their importance for the characterization of AD diseases status (Fig. 1).

Table 1

List of the 75 common proteins measured in plasma and CSF in ADNI1

S. No.	Protein	Symbol	Protein	Symbol	Protein	Symbol
1	AXL Receptor Tyrosine Kinase (ng/mL)	AXL	Fas Ligand (pg/mL)	FASLG	Neutrophil Gelatinase-Associated Lipocalin (ng/mL)	LCN2
2	Adiponectin (ug/mL)	ADIPOQ	Fatty Acid-Binding Protein, heart (ng/mL)	FABP3	Osteopontin (ng/mL)	SPP1
3	Alpha-1-Microglobulin (ug/mL)	AMBP	Ferritin (ng/mL)	FTL	Pancreatic Polypeptide (pg/mL)	PPY
4	Alpha-2-Macroglobulin (mg/mL)	A2M	Fibrinogen (mg/mL)	FGA	Placenta Growth Factor (pg/mL)	PGF
5	Alpha-1-Antitrypsin (mg/mL)	SERPINA1	Fibroblast Growth Factor 4 (pg/mL)	FGF4	Plasminogen Activator Inhibitor 1 (ng/mL)	SERPINE1
6	Agouti-Related Protein (pg/mL)	AGRP	Follicle-Stimulating Hormone (mIU/mL)	FSHB	Pregnancy-Associated Plasma Protein A (mIU/mL)	PAPPA
7	Angiotensin-Converting Enzyme (ng/mL)	ACE	Heparin-Binding EGF-Like Growth Factor (pg/mL)	HBEGF	Prolactin (ng/mL)	PRL
8	Angiopoietin-2 (ng/mL)	ANGPT2	Hepatocyte Growth Factor (ng/mL)	HGF	Prostatic Acid Phosphatase (ng/mL)	ACPP
9	Apolipoprotein A-I (mg/mL)	APOA1	Immunoglobulin A (mg/mL)	IGAD1	Resistin (ng/mL)	RETN
10	Apolipoprotein C-III (ug/mL)	APOC3	Insulin-like Growth Factor-Binding Protein-2 (ng/mL)	IGFBP2	Serum Amyloid P-Component (ug/mL)	APCS
11	Apolipoprotein D (ug/mL)	APOD	Intercellular Adhesion Molecule 1 (ng/mL)	ICAM1	Serum Glutamic Oxaloacetic Transaminase (ug/mL)	GOT1
12	Apolipoprotein E (ug/mL)	APOE	Interferon gamma Induced Protein 10 (pg/mL)	CXCL10	Sex Hormone-Binding Globulin (nmol/L)	SHBG
13	Apolipoprotein H (ug/mL)	APOH	Interleukin-16 (pg/mL)	IL16	Sortilin (ng/mL)	SORT1
14	Apolipoprotein(a) (ug/mL)	LPA	Interleukin-3 (ng/mL)	IL3	Stem Cell Factor (pg/mL)	KITLG
15	Beta-2-Microglobulin (ug/mL)	B2M	Interleukin-6 receptor (ng/mL)	IL6R	T Lymphocyte-Secreted Protein I-309 (pg/mL)	CCL1
16	C-Reactive Protein (ug/mL)	CRP	Interleukin-8 (pg/mL)	IL8	T-Cell-Specific Protein RANTES (ng/mL)	CCL5
17	CD 40 antigen (ng/mL)	CD40	Leptin (ng/mL)	LEP	TNF-Related Apoptosis-Inducing Ligand Receptor 3 (ng/mL)	TNFRSF10C
18	Calcitonin (pg/mL)	CALCA	Macrophage Colony-Stimulating Factor 1 (ng/mL)	CSF1	Thrombomodulin (ng/mL)	THBD
19	Cancer Antigen 19-9 (U/mL)	NAA15	Macrophage Inflammatory Protein-1 beta (pg/mL)	CCL3	Thyroxine-Binding Globulin (ug/mL)	SERPINA7
20	Chemokine CC-4 (ng/mL)	CCL16	Macrophage Migration Inhibitory Factor (ng/mL)	MIF	Tissue Inhibitor of Metalloproteinases 1 (ng/mL)	TIMP1
21	Chromogranin-A (ng/mL)	CHGA	Matrix Metalloproteinase-2 (ng/mL)	MMP2	Trefoil Factor 3 (ug/mL)	TFF3
22	Clusterin (ug/ml)	CLU	Monocyte Chemotactic Protein 1 (pg/mL)	CCL2	Tumor Necrosis Factor Receptor 2 (ng/mL)	TNFRSF14
23	Complement C3 (mg/mL)	C3	Monocyte Chemotactic Protein 2 (pg/mL)	CCL8	Vascular Cell Adhesion Molecule-1 (ng/mL)	VCAM1
24	Cortisol (ng/mL)	CRTL	Monokine Induced by Gamma Interferon (pg/mL)	CXCL9	Vascular Endothelial Growth Factor (pg/mL)	VEGFA
25	Cystatin-C (ng/mL)	CST3	Myoglobin (ng/mL)	MB	von Willebrand Factor (ug/mL)	VWF

Statistical analyses

We compared the baseline characteristics of study participants between three diagnostic group CN, MCI, and AD using one way analysis of variance for continuous factors and Pearson’s chi-square test for categorical factors. To define the initial set of significant proteins among all 75 proteins in plasma and CSF first we used the three logistic regression models, Model 1: single protein only (unadjusted), Model 2: single protein adjusted for age, sex and education, and Model 3: Model 2 plus APOE ϵ4 genotype. Then we selected the proteins which have p-value <0.1 in three models with respect to each diagnostic group CN versus MCI, CN versus AD, and MCI versus AD. Second, we used least absolute shrinkage and selection operator (LASSO) [33, 34] method to identify the variable importance for disease diagnosis in an iterative resampling of training-and-test based variable selection and modeling approach. We chose training and test samples using 80% and 20% proportions, respectively, based on stratified random sampling. LASSO shrinks the regression coefficients towards zero by penalizing the regression model with a penalty term called L1-norm. L1 penalty subtract a multiple (of the sum of the absolute coefficients from log likelihood, thus setting some regression coefficients to zero. The LASSO regression analysis was conducted with 10-fold cross-validation to optimize tuning parameter and select the top proteins from the training data set. In third step, Random Forest (RF) [35] algorithm was used to check the performance of top proteins in differentiating CN versus MCI, CN versus AD, and MCI versus AD individuals. We measured final model performance based on AUC, sensitivity, specificity, and accuracy. Average model performances were calculated based on test data (20%) over 500 resampling. In RF algorithm, sensitivity, specificity, and accuracy used optimal cut-off value for the prediction of case and control cases based on the minimization of the Euclidian distance between the receiver operating curve (ROC) and the (0, 1) point [36, 37]. Both LASSO and RF regression were used through R packages glmnet and randomForest, respectively. All statistical analyses were performed using R software version 4.2.1.

RESULTS

Sample characteristics

The baseline characteristics of the known risk factors in ADNI1 samples (257 total, 46 CN, 143 MCI, and 68 AD) were described in Table 2. Participants had an average [mean (SD)] age and education years 74.94 (7.10) and 15.69 (2.95), respectively, and 97 (37.74%) participants were females. There were no significant differences in age and educational attainment among the three cognitive impairment status (CN, MCI, and AD). Sex had a significant difference among CN, MCI, and AD individuals (p = 0.049) with higher percentage of male individuals. There was a significant difference between APOE ϵ4 genotype status and cognitive impairment status p < 0.001. This data set includes no APOE ϵ4 allele (n = 128), 1 APOE ϵ4 allele (n = 96), and 2 APOE ϵ4 allele (n = 33) (Table 2). AD markers CSF p-tau and Aβ₄₂ mean (SD) 30.58 (15.43) and 878.11 (456.17), respectively, were significantly associated (p < 0.001) with cognitive impairment status. As expected, the CN participants had a higher average value of Aβ₄₂, and AD participants had higher value of p-tau.

Table 2

Baseline characteristics of known risk factors in ADNI samples

Characteristic¹	Overall (n = 257)	CN (n = 46)	MCI (n = 143)	AD (n = 68)	p ²
Age	74.94 (7.10)	75.43 (5.93)	74.77 (7.20)	74.98 (7.66)	0.86
Education	15.69 (2.95)	15.52 (2.83)	16.02 (2.93)	15.12 (2.99)	0.10
Sex					0.049
Female	97 (37.74)	23 (50.00)	45 (31.47)	29 (42.65)
Male	160 (62.26)	23 (50.00)	98 (68.53)	39 (57.35)
APOEϵ4					<0.001
0	128 (49.81)	42 (91.30)	66 (46.15)	20 (29.41)
1	96 (37.35)	4 (8.70)	59 (41.26)	33 (48.53)
2	33 (12.84)	0 (0.00)	18 (12.59)	15 (22.06)
AD markers
CSF p-tau	30.61 (14.14)	19.88 (6.32)	31.73 (13.67)	35.51 (15.34)	<0.001
CSF Aβ₄₂	859.65 (450.70)	1,453.50 (271.84)	786.18 (397.57)	612.43 (275.37)	<0.001

CN, cognitive normal; MCI, mild cognitive impairment; AD, Alzheimer’s disease; CSF, cerebrospinal fluid. ¹Mean (SD)/n (%), ²One-way ANOVA/Pearson’s Chi-squared test.

Top proteins selection

All participants had the same protein measurements from both plasma and CSF samples. Among 75 proteins, at the first step, we selected the shorter panels of significant proteins (p < 0.1) based on the simple/multiple logistic regression for Model 1, Model 2, and Model 3 across the three groups CN versus MCI, CN versus AD, and MCI versus AD (Table 3). Selected proteins are listed in Supplementary Figures 1–6 in descending order of variable importance. After randomizing the whole data set into training (80%) and test (20%) sets through stratified random sampling, a LASSO logistic regression feature selection method was built with the training set using all the selected proteins in the first step. A 10-fold cross-validation procedure was performed to optimize penalization and model weight parameters for each model after dividing the training data randomly into ten equal cross-validation sets. The model’s performance was evaluated based on the never-before-seen test data sets, and the whole procedure was repeated over 500 resampling. We calculated the average variable importance (Supplementary Figures 1–6) through vip() function using the R package vip and AUC (Table 4) based on 500 samples.

Table 3

Numbers of significant proteins (p < 0.1) in each group

Sample groups (n = 257)		Model 1	Model 2	Model 3
Plasma	CN versus MCI	23	25	22
	CN versus AD	15	18	21
	MCI versus AD	23	23	26
CSF	CN versus MCI	14	15	13
	CN versus AD	18	18	15
	MCI versus AD	11	15	16

CN, cognitive normal; MCI, mild cognitive impairment; AD, Alzheimer’s disease; CSF, cerebrospinal fluid; Model 1: unadjusted model, Model 2: adjusted for age, sex and education, Model 3: Model 2 plus APOE ϵ4.

Table 4

AUC based on LASSO logistic regression in each group for three models

Sample groups (n = 257)		Model 1	Model 2	Model 3
Plasma	CN versus MCI	0.89	0.88	0.91
	CN versus AD	0.92	0.93	0.92
	MCI versus AD	0.88	0.88	0.89
CSF	CN versus MCI	0.82	0.81	0.87
	CN versus AD	0.88	0.88	0.92
	MCI versus AD	0.63	0.63	0.67

From Table 4 it was observed that plasma proteins performed better than the CSF proteins (approximately AUC 88% to 93%) across three models (Model 1: proteins only (unadjusted), Model 2: adjusted for age, sex, education and Model 3: Model 2 plus APOE ϵ4) in each diagnostic group, especially in distinguishing MCI from AD. Further we picked a subset of top proteins based on their importance in LASSO logistic regression (Supplementary Figures 1–6). The top 3 proteins were more consistent in at least two models across each comparison regardless of their order. So, we chose the top three proteins in each group (CN versus MCI, CN versus AD, and MCI versus AD), and dropped those who were not common in at least two models and then we grouped them together. In this way, we have generated 6 and 7 proteins from the plasma and CSF for model building, respectively (Table 5).

Table 5

Top proteins selected from LASSO regression.

		Model 1	Model 2	Model 3	Common^#	Combined
Plasma	CN versus MCI	APOE	APOE	C3	APOE
		AMBP	AMBP	AMBP	AMBP	APOE
		C3	C3	APOE	C3	AMBP
	CN versus AD	IL16	IL16	IL16	IL16	C3
		APOE	APOE	APOE	APOE	IL16
		VEGFA	TIMP1	ANGPT2		IGFBP2
	MCI versus AD	IGFBP2	IGFBP2	IGFBP2	IGFBP2	APOD
		APOD	APOD	APOD	APOD
		CRTL	C3	C3	C3
CSF	CN versus MCI	VEGFA	VEGFA	VEGFA	VEGFA
		HGF	HGF	FGF4	HGF	VEGFA
		PRL	PRL	HGF	PRL	HGF
	CN versus AD	VEGFA	VEGFA	PRL	VEGFA	PRL
		FGF4	FABP3	VEGFA	FABP3	FABP3
		FABP3	FGF4	FABP3	FGF4	FGF4
	MCI versus AD	CD40	CD40	CD40	CD40	CD40
		RETN	PRL	PRL	PRL	RETN
		PRL	RETN	SERPINE1	RETN

Final model selection

Next, we applied the RF algorithm to check the performance of the final three plasma models based on 6 proteins (APOE, AMBP, C3, IL16, IGFBP2, APOD) as well as three CSF models based on 7 proteins (VEGFA, HGF, PRL, FABP3, FGF4, CD40, RETN). These models were compared with typical AD biomarkers (CSF p-tau and Aβ₄₂) models (Model 1: CSF p-tau and Aβ₄₂ only (unadjusted), Model 2: CSF p-tau and Aβ₄₂ adjusted for age, sex, education, and Model 3: Model 2 plus APOE ϵ4). We calculated the average value of AUC, accuracy, sensitivity, and specificity over 500 resampling (Fig. 2 and Supplementary Figures 7–9). Plasma proteins had no strong correlation correlations; however, some CSF proteins are correlated (Fig. 3).

Fig. 2

Mean AUC for both plasma and CSF (n = 257) over 500 resampling using Random Forest algorithm based on test data. CN, cognitive normal; MCI, mild cognitive impairment; AD, Alzheimer’s disease; CSF, cerebrospinal fluid; Model 1: unadjusted model, Model 2: adjusted for age, sex and education, Model 3: Model 2 plus APOE ϵ4.

Fig. 3

Correlation plots of the selected proteins panels in plasma and cerebrospinal fluid (CSF) (n = 257) samples.

Plasma proteins outperformance of CSF typical AD biomarkers for MCI

The AUCs for 6 plasma proteins model based on RF algorithm were calculated for each disease group [CN versus MCI: 0.86 (model 1), 0.85 (model 2), 0.89 (model 3), CN versus AD: 0.85 (model 1), 0.86 (model 2), 0.91 (model 3), and MCI versus AD: 0.76 (model 1), 0.75 (model 2), 0.75 (model 3)] (Fig. 2). In summary, 6 plasma proteins panel outperformed the CSF typical AD biomarkers to differentiate MCI from AD and were equivalent to differentiate MCI from CN [AUCs for CSF p-tau and Aβ₄₂, CN versus MCI: 0.86 (model 1), 0.89 (model 2), 0.89 (model 3), CN versus AD: 0.97 (model 1), 0.98 (model 2), 0.97 (model 3), and MCI versus AD: 0.52 (model 1), 0.54 (model 2), 0.56 (model 3)] (Fig. 2). In addition, these models were independent from known risks factors age, sex, education, and APOE ϵ4 genotype to differentiate MCI and AD patients.

Plasma proteins outperformance of CSF proteins across diagnostic groups

We also observed that 6 plasma proteins panel outperformed the 7 CSF proteins panel to distinguishing the CN, MCI, and AD groups. The AUCs for 7 CSF proteins model [CN versus MCI: 0.77 (model 1), 0.78 (model 2), 0.85 (model 3), CN versus AD: 0.82 (model 1), 0.83 (model 2), 0.89 (model 3) and MCI versus AD: 0.59 (model 1), 0.59 (model 2), 0.56 (model 3)] were also estimated (Fig. 2). It was determined that the performance of CSF proteins as well as p-tau and Aβ₄₂ models were better in CN versus AD than the CN versus MCI and MCI versus AD.

DISCUSSION

In this study, we used LASSO analyses and compared the performance of 75 commonly measured proteins in plasma and CSF for accuracy of disease classification at baseline. For the primary analyses it is observed that more plasma proteins were significantly associated with CN versus MCI, CN versus AD, and MCI versus AD classification (Table 3). Our present study results suggest that 6 protein candidates in plasma (APOE, AMBP, C3, IL16, IGFBP2, APOD) and 7 candidate protein in CSF (VEGFA, HGF, PRL, FABP3, FGF4, CD40, RETN) were important predictors. The diagnostic performance of these candidates was estimated and compared with conventional AD biomarkers namely CSF p-tau and Aβ₄₂ for MCI.

One of the key findings of our study is that the 6 plasma proteins panel performs better than CSF p-tau and Aβ₄₂ in classifying MCI versus AD subjects and resulted in good accuracy as CSF p-tau and Aβ₄₂ in classifying CN versus MCI subjects. Further, combining plasma proteins panel with age, sex, education, and APOE ϵ4 had no effect on the model’s performance (Fig. 2 and Supplementary Figure 7–9). CSF p-tau and Aβ₄₂ were measured by fully automated Roche Elecsys but we also consider CSF p-tau and Aβ₄₂ measured by AlzBio3 immunoassay to compare our results and we observed that still results are same (Supplementary Figure 12). ATN is the primary pathological trait that distinguishes AD from other diseases, and ATN biomarkers are increasingly used to support an AD continuum diagnosis. However, in clinical setting, the classification of pre-symptomatic subjects on the AD continuum is readily accepted due to the absence of disease-modifying therapies available to this disease stage. Further, elevated Aβ levels can be commonly found in cognitively intact older adults as well as in those with non-AD dementias or mixed dementias [8, 38]. Although Aβ and tau biomarkers have been reported to predict incident AD dementia in those with subjective cognitive decline, individualized risk modeling remains difficult [8]. To assess Aβ plaque deposition, typically CSF assays or amyloid PET imaging are used. Even though these tests provide valid results on Aβ deposition status and their use in measuring biomarkers is mentioned in the diagnostic criteria for AD [39], a person who tests positive for Aβ by PET may require additional tests to determine whether the amyloidosis is caused by AD or advanced age. In addition, the lumbar puncture procedure used to obtain CSF samples from patients is invasive and patients are frequently resistant to this procedure. However, due to low cost and easy measurements, these plasma proteins could be used as a screening test for MCI or AD. If a positive result is shown, more expensive and invasive tests like CSF and PET scan for AD will be suggested to patients.

We found that the top proteins in plasma were largely different from the top proteins in CSF and outperformed from CSF for determining MCI (Fig. 2 and Supplementary Figures 7–9), suggesting that peripheral proinflammatory factors increase AD risk at an early stage via different molecular pathways from the ones in CSF. Several studies demonstrated that combinations of plasma proteins were useful for AD classification and prediction. A recent study identified numerous pathway-specific plasma proteins that could be used as peripheral biomarkers in the early stages of AD and related dementias. In this study, 32 dementia-associated plasma proteins were revealed to be involved in proteostasis, immunology, synaptic function, and extracellular matrix architecture [40]. A combination of 18 plasma signaling and inflammatory proteins (ANG-2, CCL5, CCL7, CCL15, CCL18, CXCL8, EGF, G-CSF, GDNF, CAM-1, IGFBP-6, IL-1α, IL-3, L-11, M-CSF, PDGF-BB, TNF-α, and TRAIL-R4) can be used to differentiate CN versus AD subjects with an approximate accuracy of 90% and to predict the AD progression from MCI 2–6 years later [19, 20]. A model of ten plasma proteins (MTDH, ADIPOQ, APOB, TF, CA1, C9, APOA4, RBP4, F13A1, and FGA) derived by multiple reaction monitoring mass spectrometry was developed, which classified the CN versus AD and CN versus asymptomatic AD subjects with the AUCs of 81.7% and 77.7%, respectively, when combining with APOE ϵ4, model performance improved the AUCs of 87.3% and 82.6%, respectively [21]. Another study found that a set of eight plasma proteins (BDNF, AGT, IGFBP-2, OPN, cathepsin D, SAP, complement C4, and TTR) may function as a valuable diagnostic biomarker for AD in the Chinese population [22]. A combination of two plasma and four CSF proteins were used to predict mid-term AD dementia progression with sensitivity of 88% and a specificity of 70% [23]. Lower level of plasma C-reactive protein was associated with AD diagnosis and rapid progression to AD dementia [41 –43].

In the selected plasma proteins in this study only IGFBP2 overlapped with the previous studies Ray et al. [20], O’Bryant et al. [44], and Cheng et al. [22]. IGFBP2 can affect the formation of DNA, cell growth, death, and cellular uptake of glucose and amino acids by preventing IGF functions [22, 45]. The mean level of plasma IGFBP2 was significantly high in MCI than the CN and AD p < 0.01 in both comparisons (Supplementary Figure 10). The significantly highly expressed plasma proteins in healthy population were APOE (p < 0.001) and IL16 (p < 0.01). APOE is a polymorphic lipoprotein that is a primary cholesterol transporter in the brain and it is generated and secreted from a number of tissues and cell types and is abundant in interstitial fluid, lymph, and plasma [46, 47]. IL16 is a chemoattractant for specific immune cells, and its overexpression has been reported in a variety of cerebral diseases, including AD [48]. Plasma AMBP differentially expressed in CN versus MCI (p < 0.01) and MCI versus AD (p < 0.001) and plasma C3 differentially expressed in CN versus MCI (p < 0.01) and MCI versus AD (p < 0.001). AMBP is a precursor for bikunin, a lipid transporter that has also been linked to inflammation control and neuroprotection and it is primarily synthesized in the liver and is upregulated in response to heme, reactive oxygen species and to the pro-inflammatory cytokines [49, 50]. C3, a large molecular weight protein (185 kDa), is a key component of the innate immune system, providing a primary host mechanism for the recognition and clearance of pathogenic microbes along with other complement proteins [51]. Plasma APOD is significantly decrease in MCI than CN and AD (p < 0.05). APOD is broadly expressed in humans, unlike to other apolipoproteins, which are mostly generated by the liver, implying fundamental cellular roles [52]. It is involved in oxidative stress, inflammation, and small hydrophobic molecule transport. APOD has been identified as the gene that is most upregulated with age, and its expression increases in a variety of illnesses, most notably Parkinson’s disease and AD [53].

Both plasma and CSF proteins analyzed by same technology as discussed above, and it is noticed that 45 proteins out of 75 were correlated (Supplementary Table 1) in both plasma and CSF. Four proteins in the selected 6 plasma proteins (APOE, AMBP, C3, IL16, IGFBP2, APOD) were significantly correlated with their CSF sample (Protein, correlation [p]: AMBP, 0.43 [5E-13], IL16, 0.2398 [1E-04], IGFBP2, 0.6609 [1.2E-33], APOD, 0.1474 [0.02]). However, these proteins were not found to be top CSF proteins for AD. Also, these proteins have significantly higher expression levels in plasma than CSF (Supplementary Figure 13). Interestingly, some of the plasma proteins including C3, IGFBP2, and APOD presented with a U shape when comparing CN, MCI, and AD (Supplementary Figure 10). Thus, these plasma proteins are not brain-specific and may have a specific influence from peripheral system on AD development and progression.

CSF protein models (Models 1, 2, and 3) performed satisfactorily only in distinct CN versus AD subjects, with AUCs of 88%, 88%, and 92% (Table 4) based on the number of proteins in each model (18, 18, and 15 (Table 3), respectively). These results were better than the previous results using the same data with model of 24 proteins have an AUC of 80% [54]. These models were not good classifier of the MCI versus AD groups AUC <70% (Table 4). Further, the performance of these models with 7 candidates protein models was also poor, with AUCs ≤60% in the MCI versus AD group. We observed that CSF p-tau and Aβ₄₂ were inadequate in differentiating MCI from AD subjects, with AUCs of 52%, 54%, and 56% (Fig. 2). However, 6 plasma candidates protein marker (Models) provide a much better AUC ≥75% for differentiating MCI from AD subjects. This implies that CSF biomarkers of either 7 candidate proteins or p-tau and Aβ₄₂ are good predictors in distinguishing the CN versus AD subjects. Mean level of 7 CSF proteins were significantly different in CN versus AD, four proteins VEGFA, HGF, FABP3 and FGF4 were significantly different in CN versus MCI and three CSF proteins VEGFA, CD40, and RETN were differentially expressed in MCI versus AD.

Although these are some promising results, our study carries a few limitations. Even though we identified plasma proteins that show potential in the classification of CN versus MCI and MCI versus AD subjects through training and test datasets, our findings are limited by the inability to validate these candidates in a separate cohort. Future research into the prognostic potential of the candidates identified here in other well-characterized prospective cohorts is required. Furthermore, we only chose 75 plasma proteins (out of 146) in order to compare the same proteins in CSF and plasma. As a result, future research may look at the entire set of plasma proteins to investigate the different set of candidate proteins in plasma for detecting AD. Finally, longitudinally monitoring these plasma proteins in the same subject along with cognitive function changes will help examine the causal role of these proteins. Using ADNI data set have lack of diversity and overall health of the cohort. Also, due to the insufficient sample size molecular PET imaging biomarkers were not included in this study. Thus, we were limited to compare the performance of the selected protein panels with CSF p-tau and Aβ₄₂.

In conclusion, we observed that plasma candidate proteins performed well in differentiating MCI from both CN and AD subjects. However, CSF candidate proteins and Aβ₄₂ and p-tau have a higher accuracy in distinguishing CN from AD subjects. As a result of these findings, plasma candidate marker can be considered an effective diagnostic tool for the early detection and monitoring progression of AD symptoms and later distinguishing AD from MCI. Our study also supports that peripheral inflammation increases AD risk at an early stage.

Footnotes

ACKNOWLEDGMENTS

Data collection and sharing for this project was funded by the Alzheimer’s Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012).

ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer’s Association; Alzheimer’s Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (https://www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer’s Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California. Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of the ADNI and/or provided data but did not participate in the analysis or writing of this report. A complete listing of ADNI investigators can be found at .

FUNDING

This study was supported by National Institute on Aging grants U19-AG068753, RF1-G057519, and R01-AG048927.

CONFLICT OF INTEREST

Rhoda Au is an Editorial Board Member of this journal but was not involved in the peer-review process nor had access to any information regarding its peer-review.

All other authors have no conflict of interest to report.

DATA AVAILABILITY

The data used in the preparation of this study were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) phase 1 (ADNI1) database ().

The supplementary material is available in the electronic version of this article: .

References

DeTure

, Dickson

(2019) The neuropathological diagnosis of Alzheimer’s disease, Mol Neurodegener 14, 32.

Alzheimer’s Association(2010) 2010 Alzheimer’s disease facts and figures. Alzheimers Dement 6, 158–194.

Schneider

, Hampel

, Buerger

(2009) Biological marker candidates of Alzheimer’s disease in blood, plasma, and serum. CNS Neurosci Ther 15, 358–374.

Arnerić

, Batrla-Utermann

, Beckett

, Bittner

, Blennow

, Carter

, Dean

, Engelborghs

, Genius

, Gordon

(2017) Cerebrospinal fluid biomarkers for Alzheimer’s disease: A view of the regulatory science qualification landscape from the coalition against major diseases CSF biomarker team. J Alzheimers Dis 55, 19–35.

Rajmohan

, Reddy

(2017) Amyloid-beta and phosphorylated tau accumulations cause abnormalities at synapses of Alzheimer’s disease neurons. J Alzheimers Dis 57, 975–999.

Scheltens

, Blennow

, Breteler

, De Strooper

, Frisoni

, Salloway

, Van der Flier

(2016) Alzheimer’s disease. Lancet 388, 505–517.

Blennow

, de Leon

, Zetterberg

(2006) Alzheimer’s disease. Lancet 368, 387–403.

Khoury

, Ghossoub

(2019) Diagnostic biomarkers of Alzheimer’s disease: A state-of-the-art review. Biomark Neuropsychiatry 1, 100005.

Blennow

, Hampel

, Weiner

, Zetterberg

(2010) Cerebrospinal fluid and plasma biomarkers in Alzheimer disease. Nat Rev Neurol 6, 131–144.

10.

Bouwman

, Frisoni

, Johnson

, Chen

, Engelborghs

, Ikeuchi

, Paquet

, Ritchie

, Bozeat

, Quevenco

(2022) Clinical application of CSF biomarkers for Alzheimer’s disease: From rationale to ratios, Alzheimers Dement (Amst) 14, e12314.

11.

Jack

Jr , Bennett

, Blennow

, Carrillo

, Dunn

, Haeberlein

, Holtzman

, Jagust

, Jessen

, Karlawish

(2018) NIA-AA research framework: Toward a biological definition of Alzheimer’s disease. Alzheimers Dement 14, 535–562.

12.

Grothe

, Moscoso

, Ashton

, Karikari

, Lantero-Rodriguez

, Snellman

, Zetterberg

, Blennow

, Schöll

, Alzheimer’s Disease Neuroimaging Initiative (2021) Associations of fully automated CSF and novel plasma biomarkers with Alzheimer disease neuropathology at autopsy, Neurology 97, e1229–e1242.

13.

Schöll

, Maass

, Mattsson

, Ashton

, Blennow

, Zetterberg

, Jagust

(2019) Biomarkers for tau pathology. Mol Cell Neurosci 97, 18–33.

14.

Rehiman

, Lim

, Neoh

, Majeed

ABA

, Chin

A-V

, Tan

, Kamaruzzaman

, Ramasamy

(2020) Proteomics as a reliable approach for discovery of blood-based Alzheimer’s disease biomarkers: A systematic review and meta-analysis. Ageing Res Rev 60, 101066.

15.

Cullen

, Leuzy

, Janelidze

, Palmqvist

, Svenningsson

, Stomrud

, Dage

, Mattsson-Carlgren

, Hansson

(2021) Plasma biomarkers of Alzheimer’s disease improve prediction of cognitive decline in cognitively unimpaired elderly populations. Nat Commun 12, 3555.

16.

Palmqvist

, Tideman

, Cullen

, Zetterberg

, Blennow

, Alzheimer’s Disease Neuroimaging I Dage

, Stomrud

, Janelidze

, Mattsson-Carlgren

, Hansson

(2021) Prediction of future Alzheimer’s disease dementia using plasma phospho-tau combined with other accessible measures. Nat Med 27, 1034–1042.

17.

Mattsson-Carlgren

, Salvado

, Ashton

, Tideman

, Stomrud

, Zetterberg

, Ossenkoppele

, Betthauser

, Cody

, Jonaitis

, Langhough

, Palmqvist

, Blennow

, Janelidze

, Johnson

, Hansson

(2023) Prediction of longitudinal cognitive decline in preclinical Alzheimer disease using plasma biomarkers. JAMA Neurol 80, 360–369.

18.

Klyucherev

, Olszewski

, Shalimova

, Chubarev

, Tarasov

, Attwood

, Syvänen

, Schiöth

(2022) Advances in the development of new biomarkers for Alzheimer’s disease. Transl Neurodegener 11, 25.

19.

Britschgi

, Wyss-Coray

(2009) Blood protein signature for the early diagnosis of Alzheimer disease. Arch Neurol 66, 161–165.

20.

Ray

, Britschgi

, Herbert

, Takeda-Uchimura

, Boxer

, Blennow

, Friedman

, Galasko

, Jutel

, Karydas

, Kaye

, Leszek

, Miller

, Minthon

, Quinn

, Rabinovici

, Robinson

, Sabbagh

, So

, Sparks

, Tabaton

, Tinklenberg

, Yesavage

, Tibshirani

, Wyss-Coray

(2007) Classification and prediction of clinical Alzheimer’s diagnosis based on plasma signaling proteins. Nat Med 13, 1359–1362.

21.

Kim

, Kim

, Son

, Lee

, Yeo

, Choi

, Kim

, Lee

, Kim

(2022) Plasma protein biomarker model for screening Alzheimer disease using multiple reaction monitoring-mass spectrometry. Sci Rep 12, 1282.

22.

Cheng

, Yin

, Yuan

, Jin

, Zhang

, Wang

, Liu

, Wu

, Wang

, Xiao

(2018) Blood-derived plasma protein biomarkers for Alzheimer’s disease in Han Chinese. Front Aging Neurosci 10, 414.

23.

Lehallier

, Essioux

, Gayan

, Alexandridis

, Nikolcheva

, Wyss-Coray

, Britschgi

, Alzheimer’s Disease Neuroimaging Initiative (2016) Combined plasma and cerebrospinal fluid signature for the prediction of midterm progression from mild cognitive impairment to Alzheimer disease. JAMA Neurol 73, 203–212.

24.

Tao

, Ang

TFA

, Akhter-Khan

, Itchapurapu

, Killiany

, Zhang

, Budson

, Turk

, Goldstein

, Mez

(2021) Impact of C-reactive protein on cognition and Alzheimer disease biomarkers in homozygous APOE ϵ4 carriers, Neurology 97, e1243–e1252.

25.

Huang

, Lin

, Yang

, Wang

, Cai

, Pang

, Wu

, Huang

Alzheimer’s Disease Neuroimaging Initiative (2020) A multipredictor model to predict the conversion of mild cognitive impairment to Alzheimer’s disease by using a predictive nomogram. Neuropsychopharmacology 45, 358–366.

26.

Huang

, Tao

, Ang

TFA

, Farrell

, Zhu

, Wang

, Stein

, Lunetta

, Massaro

, Mez

(2022) The impact of increasing levels of blood C-reactive protein on the inflammatory loci SPI1 and CD33 in Alzheimer’s disease. Transl Psychiatry 12, 523.

27.

Abe

, Chiba

, Hattori

, Tamazawa

, Yoshimi

, Katsuse

, Suda

, Alzheimer’s Disease Neuroimaging Imitative (2020) Influence of plasma cytokine levels on the conversion risk from MCI to dementia in the Alzheimer’s disease neuroimaging initiative database. J Neurol Sci 414, 116829.

28.

Tubi

, Kothapalli

, Hapenney

, Feingold

, Mack

, King

, Thompson

, Braskie

for Alzheimer’s Disease Neuroimaging Initiative (2021) Regional relationships between CSF VEGF levels and Alzheimer’s disease brain biomarkers and cognition. Neurobiol Aging 105, 241–251.

29.

Wirth

, Lange

, Huijbers

Alzheimer’s Disease Neuroimaging Initiative (2019) Plasma cortisol is associated with cerebral hypometabolism across the Alzheimer’s disease spectrum, Neurobiol Aging 84, 80–89.

30.

Aisen

, Petersen

, Donohue

, Gamst

, Raman

, Thomas

, Walter

, Trojanowski

, Shaw

, Beckett

(2010) Clinical core of the Alzheimer’s Disease Neuroimaging Initiative: Progress and plans. Alzheimers Dement 6, 239–246.

31.

Bittner

, Zetterberg

, Teunissen

, Ostlund

Jr , Militello

, Andreasson

, Hubeek

, Gibson

, Chu

, Eichenlaub

, Heiss

, Kobold

, Leinenbach

, Madin

, Manuilova

, Rabe

, Blennow

(2016) Technical performance of a novel, fully automated electrochemiluminescence immunoassay for the quantitation of beta-amyloid (1-42) in human cerebrospinal fluid. Alzheimers Dement 12, 517–526.

32.

Saykin

, Shen

, Foroud

, Potkin

, Swaminathan

, Kim

, Risacher

, Nho

, Huentelman

, Craig

(2010) Alzheimer’s Disease Neuroimaging Initiative biomarkers as quantitative phenotypes: Genetics core aims, progress, and plans. Alzheimers Dement 6, 265–273.

33.

Heinze

, Wallisch

, Dunkler

(2018) Variable selection - A review and recommendations for the practicing statistician. Biom J 60, 431–449.

34.

Utazirubanda

, Leon

, Ngom

(2021) Variable selection with group LASSO approach: Application to Cox regression with frailty model. Commun Stat Simul Comput 50, 881–901.

35.

(2012) Random forest for bioinformatics. In Ensemble Machine Learning: Methods and Applications. Springer, pp. 307-323.

36.

Perkins

, Schisterman

(2006) The inconsistency of “Optimal” cut-points using two ROC based criteria. Am J Epidemiol 163, 670–675.

37.

Unal

(2017) Defining an optimal cut-point value in ROC analysis: An alternative approach. Comput Math Methods Med 2017, 3762651.

38.

Cummings

(2019) The National Institute on Aging—Alzheimer’s Association framework on Alzheimer’s disease: Application to clinical trials. Alzheimers Dement 15, 172–178.

39.

McKhann

, Knopman

, Chertkow

, Hyman

, Jack

Jr , Kawas

, Klunk

, Koroshetz

, Manly

, Mayeux

(2011) The diagnosis of dementia 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, 263–269.

40.

Walker

, Chen

, Shi

, Yang

, Fornage

, Zhou

, Schlosser

, Surapaneni

, Grams

, Duggan

(2023) Proteomics analysis of plasma from middle-aged adults identifies protein markers of dementia risk in later life, Sci Transl Med 15, eadf5681.

41.

Yarchoan

, Louneva

, Xie

, Swenson

, Hu

, Soares

, Trojanowski

, Lee

VM-Y

, Kling

, Shaw

(2013) Association of plasma C-reactive protein levels with the diagnosis of Alzheimer’s disease. J Neurol Sci 333, 9–12.

42.

Fernandes

, Tábuas-Pereira

, Duro

, Lima

, Gens

, Santiago

, Durães

, Almeida

, Leitão

, Baldeiras

(2020) C-reactive protein as a predictor of mild cognitive impairment conversion into Alzheimer’s disease dementia. Exp Gerontol 138, 111004.

43.

O’Bryant

, Waring

, Hobson

, Hall

, Moore

, Bottiglieri

, Massman

, Diaz-Arrastia

(2010) Decreased C-reactive protein levels in Alzheimer disease. J Geriatr Psychiatry Neurol 23, 49–53.

44.

O’Bryant

, Xiao

, Barber

, Reisch

, Doody

, Fairchild

, Adams

, Waring

, Diaz-Arrastia

Texas Alzheimer’s Research Consortium (2010) A serum protein-based algorithm for the detection of Alzheimer disease. Arch Neurol 67, 1077–1081.

45.

Reyer

, Schindler

, Ohde

, Walz

, Kunze

, Tuchscherer

, Wirthgen

, Brenmoehl

, Hoeflich

(2015) The RGD sequence present in IGFBP-2 is required for reduced glucose clearance after oral glucose administration in female transgenic mice, Am J Physiol Endocrinol Metab 309, E409–E417.

46.

Williams

, Borchelt

, Chakrabarty

(2020) Therapeutic approaches targeting Apolipoprotein E function in Alzheimer’s disease. Mol Neurodegener 15, 8.

47.

Huang

, Mahley

(2014) Apolipoprotein E: Structure and function in lipid metabolism, neurobiology, and Alzheimer’s diseases. Neurobiol Dis 72, 3–12.

48.

Guo

L-H

, Alexopoulos

, Wagenpfeil

, Kurz

, Perneczky

Alzheimer’s Disease Neuroimaging Initiative (2013) Plasma proteomics for the identification of Alzheimer’s disease. Alzheimer Dis Assoc Disord 27, 337–342.

49.

Ahmed

, Ahmed

, Rådegran

(2021) Plasma tumour and metabolism related biomarkers AMBP, LPL and Glyoxalase I differentiate heart failure with preserved ejection fraction with pulmonary hypertension from pulmonary arterial hypertension. Int J Cardiol 345, 68–76.

50.

Higginbotham

, Ping

, Dammer

, Duong

, Zhou

, Gearing

, Hurst

, Glass

, Factor

, Johnson

(2020) Integrated proteomics reveals brain-based cerebrospinal fluid biomarkers in asymptomatic and symptomatic Alzheimer’s disease, Sci Adv 6, eaaz9360.

51.

Delanghe

, Speeckaert

(2022) Host polymorphisms and COVID-19 infection. Adv Clin Chem 107, 41–77.

52.

Muffat

, Walker

(2010) Apolipoprotein D: An overview of its role in aging and age-related diseases. Cell Cycle 9, 269–273.

53.

Fyfe-Desmarais

, Desmarais

, Rassart

, Mounier

(2023) Apolipoprotein D in oxidative stress and inflammation. Antioxidants 12, 1027.

54.

Khan

, Aguilar

, Kiddle

, Doyle

, Thambisetty

, Muehlboeck

, Sattlecker

, Newhouse

, Lovestone

, Dobson

(2015) A subset of cerebrospinal fluid proteins from a multi-analyte panel associated with brain atrophy, disease classification and prediction in Alzheimer’s disease, PLoS One 10, e0134368.