Multi-proteomic analyses of 5xFAD mice reveal new molecular signatures of early-stage Alzheimer's disease

Abstract

An early diagnosis of Alzheimer's disease is crucial as treatment efficacy is limited to the early stages. However, the current diagnostic methods are limited to mid or later stages of disease development owing to the limitations of clinical examinations and amyloid plaque imaging. Therefore, this study aimed to identify molecular signatures including blood plasma extracellular vesicle biomarker proteins associated with Alzheimer's disease to aid early-stage diagnosis. The hippocampus, cortex, and blood plasma extracellular vesicles of 3- and 6-month-old 5xFAD mice were analyzed using quantitative proteomics. Subsequent bioinformatics and biochemical analyses were performed to compare the molecular signatures between wild type and 5xFAD mice across different brain regions and age groups to elucidate disease pathology. There was a unique signature of significantly altered proteins in the hippocampal and cortical proteomes of 3- and 6-month-old mice. The plasma extracellular vesicle proteomes exhibited distinct informatic features compared with the other proteomes. Furthermore, the regulation of several canonical pathways (including phosphatidylinositol 3-kinase/protein kinase B signaling) differed between the hippocampus and cortex. Twelve potential biomarkers for the detection of early-stage Alzheimer's disease were identified and validated using plasma extracellular vesicles from stage-divided patients. Finally, integrin α-IIb, creatine kinase M-type, filamin C, glutamine γ-glutamyltransferase 2, and lysosomal α-mannosidase were selected as distinguishing biomarkers for healthy individuals and early-stage Alzheimer's disease patients using machine learning modeling with approximately 79% accuracy. Our study identified novel early-stage molecular signatures associated with the progression of Alzheimer's disease, thereby providing novel insights into its pathogenesis.

Keywords: Alzheimer's disease; biomarker; early‐stage Alzheimer's disease; extracellular vesicle; machine learning; proteomics.

FIGURE 1

Preparation of multi‐proteomes from wild type (WT) and 5xFAD mice. (a) Study workflow. (b) Stained images using Thioflavin‐S and anti‐Aβ_1‐16 (clone 6E10) in the medial prefrontal cortex (mPFC) and hippocampus (HPC) of WT and 5xFAD mice. Scale bar = 100 μm. (c) Densitometric graphs of the immunostained images in (b). Data are presented as mean ± standard error (SE) (n = 3/group). *p < 0.03, **p < 0.01, and ***p < 0.001 in analysis of variance (ANOVA) with Bonferroni's multiple comparisons. (d) Western blotting of amyloid precursor protein (APP) in brain lysates of WT and 5xFAD mice. (e) Densitometric graphs of western blotting in d. Data are presented as mean ± SE (n = 3/group). *p < 0.03 in ANOVA with Bonferroni's multiple comparison test. (f) Nanovesicle‐tracking analysis of plasma extracellular vesicles (EVs). NanoSight LM10 was used to estimate plasma EV size. (g) Western blotting of plasma EV marker proteins. Plasma EVs from WT and 5xFAD mice were electrophoresed and blotted using anti‐cluster of differentiation (CD) 9, anti‐CD63, anti‐CD81, and anti‐actin antibodies.

FIGURE 2

Comparative proteomic analysis between 3‐month‐old WT and 5xFAD mice. (a) Venn diagram of the identified proteins. (b) Enrichment analyses of Gene Ontology (GO)‐based functional annotations in the hippocampus, cortex, and plasma of extracellular vesicles (EVs). (c) Comparative canonical pathway analyses among 3‐month‐old proteomes using Ingenuity Pathway Analysis (IPA). Orange and blue indicate canonical pathways with a positive or negative Z‐score, respectively, for pathway activation. Analysis parameters were a z‐score cut‐off of 0.5 and a −log (p‐value) value of >1.3. (d) Comparative analysis of the top three pathways in the three proteomes. Proteins involved in the top three IPA pathways were identified. (e) Significantly altered proteins in each proteome. Orange and blue indicate a log₂ fold increase and decrease, respectively.

FIGURE 3

Comparative proteomic analysis between 6‐month‐old WT and 5xFAD mice. (a) Enrichment analysis of GO‐based functional annotation of the hippocampus, cortex, and plasma EVs. (b) Comparative canonical pathway analyses of 6‐month‐old proteomes using IPA. Orange and blue indicate canonical pathways with a positive or negative Z‐score, respectively, for pathway activation. Analysis parameters were a z‐score cut‐off of 0.5 and a −log (p‐value) value of >1.3. (c) Comparative analysis of the top three pathways in the three proteomes. The proteins involved in the top three IPA pathways were extracted. The top pathways were synaptogenic signaling, EIF2 signaling, mitochondrial dysfunction, acute phase response, LXR/RXR activation, and FXR/RXR activation. (d) Significantly altered proteins in each proteome between 6‐month‐old WT and 5xFAD mice. Orange and blue indicate log₂ fold increase and decrease, respectively. (e) Comparative analysis of activated and deactivated canonical pathways between 6‐month‐old and 3‐month‐old mice. “Activated” and “Deactivated” indicate activated and deactivated canonical pathways, respectively, in 5xFAD mice analyzed using IPA. “Mixed” indicates a pathway demonstrating mixed directions of activation and deactivation for each proteome.

FIGURE 4

Comparative bioinformatics analysis between 3‐ and 6‐month‐old 5xFAD mice. (a) Comparative canonical pathway analysis of the three proteomes using IPA. Orange and blue indicate canonical pathways with a positive or negative Z‐score, respectively, for pathway activation. Analysis parameters were a z‐score cut‐off value of 0.5 and a −log (p‐value) value of >1.3. (b) IPA of phosphoinositide 3‐kinase/protein kinase B (PI3K/Akt) signaling in the hippocampal proteome. Red and green indicate upregulation and downregulation, respectively, in the 6‐month‐old proteome, compared with the 3‐month‐old proteome. Yellow arrows indicate contrasting expression patterns of signaling proteins in the hippocampal and cortical proteomes. (c) Western blotting of Akt (S473) phosphorylation in the hippocampi and cortices of 3‐ and 6‐month‐old 5xFAD mouse brains. The tissue lysates were electrophoresed and blotted with each antibody. Actin was used as a loading control. (d) Densitometric graphs of western blotting in “c”. Data are presented as mean ± SE (n = 3/group). *p < 0.03 in ANOVA with Bonferroni's multiple comparison test. (e) Comparative disease and biofunction analyses of the three proteomes using IPA. Orange and blue indicate canonical pathways with a positive or negative Z‐score, respectively, for pathway activation. (f) Enrichment analysis of brain pathology‐related terms among the three proteomes using IPA. Orange and blue indicate canonical pathways with a positive or negative Z‐score, respectively, for pathway activation. (g) Significantly altered proteins in the proteomes of 3‐ and 6‐month‐old 5xFAD mice. Orange and blue indicate a log₂ fold increase and decrease, respectively.

FIGURE 5

Validation of biomarker candidate proteins using plasma EVs from stage‐divided patients with AD. Western blotting of selected biomarker candidates in plasma EVs from healthy individuals and patients with early‐ and late‐stage AD. Plasma EVs were electrophoresed and blotted with antibodies. “Early” and “late” stages of AD were diagnosed using mini‐mental state examination scoring. (a) Classification according to the results of western blotting. (b) Scatterplot of class 1 proteins, including A2M, CKM, FLNA, ITGA2B, ORM2, and PLTP. (c) Scatterplot of class 2 proteins, including HP, QSOX1, and TGM2. (d) Scatter plot of class 3 proteins, including FLNC, HSP70, and MAN2B1. All results were densitometrically analyzed using ImageJ Ver 1.53 after normalization to the density of bands stained with Ponceau S. ***p < 0.001, **p < 0.01, and *p < 0.03 in one‐way ANOVA with Bonferroni's multiple comparisons (n = 39–47/group). ns, not significant.

FIGURE 6

Performance test of the proposed ML model based on the AUC‐ROC curve. Accumulated accuracy, sensitivity (SEN), specificity (SPE), and AUC‐ROC curves for (a) healthy versus early‐stage AD, (b) healthy versus late‐stage AD, and (c) early‐stage AD versus late‐stage AD. The bottom figures show the AUC‐ROC curves of the selected features.