Multi-analyte proteomic analysis identifies blood-based neuroinflammation, cerebrovascular and synaptic biomarkers in preclinical Alzheimer's disease

Abstract

Background: Blood-based biomarkers are gaining grounds for Alzheimer's disease (AD) detection. However, two key obstacles need to be addressed: the lack of methods for multi-analyte assessments and the need for markers of neuroinflammation, vascular, and synaptic dysfunction. Here, we evaluated a novel multi-analyte biomarker platform, NULISAseq CNS disease panel, a multiplex NUcleic acid-linked Immuno-Sandwich Assay (NULISA) targeting ~120 analytes, including classical AD biomarkers and key proteins defining various disease hallmarks.

Methods: The NULISAseq panel was applied to 176 plasma samples from the MYHAT-NI cohort of cognitively normal participants from an economically underserved region in Western Pennsylvania. Classical AD biomarkers, including p-tau181 p-tau217, p-tau231, GFAP, NEFL, Aβ40, and Aβ42, were also measured using Single Molecule Array (Simoa). Amyloid pathology, tau pathology, and neurodegeneration were evaluated with [11C] PiB PET, [18F]AV-1451 PET, and MRI, respectively. Linear mixed models were used to examine cross-sectional and Wilcoxon rank sum tests for longitudinal associations between NULISA biomarkers and AD pathologies. Spearman correlations were used to compare NULISA and Simoa.

Results: NULISA concurrently measured 116 plasma biomarkers with good technical performance, and good correlation with Simoa measures. Cross-sectionally, p-tau217 was the top hit to identify Aβ pathology, with age, sex, and APOE genotype-adjusted AUC of 0.930 (95%CI: 0.878-0.983). Fourteen markers were significantly decreased in Aβ-PET+ participants, including TIMP3, which regulates brain Aβ production, the neurotrophic factor BDNF, the energy metabolism marker MDH1, and several cytokines. Longitudinally, FGF2, IL4, and IL9 exhibited Aβ PET-dependent yearly increases in Aβ-PET+ participants. Markers with tau PET-dependent longitudinal changes included the microglial activation marker CHIT1, the reactive astrogliosis marker CHI3L1, the synaptic protein NPTX1, and the cerebrovascular markers PGF, PDGFRB, and VEFGA; all previously linked to AD but only reliably measured in cerebrospinal fluid. SQSTM1, the autophagosome cargo protein, exhibited a significant association with neurodegeneration status after adjusting age, sex, and APOE ε4 genotype.

Conclusions: Together, our results demonstrate the feasibility and potential of immunoassay-based multiplexing to provide a comprehensive view of AD-associated proteomic changes. Further validation of the identified inflammation, synaptic, and vascular markers will be important for establishing disease state markers in asymptomatic AD.

Keywords: NULISA with next-generation sequencing readout (NULISAseq); NUcleic acid-Linked Immuno-Sandwich Assay (NULISA); Preclinical Alzheimer’s disease; amyloid pathology; neurodegeneration; plasma biomarkers; proteomics; tau pathology.

Fig. 1:

Performance of the NULISAseq CNS disease panel. A Box plots illustrating the detectability of 116 targets in 176 plasma samples collected from 113 MYHAT-NI participants. The y-axis represents NPQ - LOD, where values >0 indicate detectability. On each box, the central mark indicates the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points not considered outliers, and the outliers are plotted individually using the '+' marker symbol. Data points were considered outliers if they were greater than q3 + 1.5 × (q3 − q1) or less than q1 − 1.5 × (q3 − q1), where q1 and q3 are the 25th and 75th percentiles of the sample data. B-C Histogram distribution of intra-plate (B) and inter-plate (C) coefficient of variations (CVs). D-E Scatterplot distributions between abundance rank and intra-plate (D) or inter-plate (E) CVs. Intra- and inter-plate CVs were calculated based on results of a pooled plasma sample (SC), measured in duplicate each in two different plates. Abundance rank was based on the mass spectrometry-estimated protein abundance in the Human Protein Atlas (downloaded on 12/24/2023). (F) Scatterplot distributions illustrating the correlation of protein levels measured using NULISAseq and Simoa method. Rho and p values were determined using Spearman rank-based correlation. Purple lines indicated the least square regression lines. NPQ, NULISA Protein Quantification; LOD, limit of detection.

Fig. 2:

Cross-sectional association of NULISAseq targets with amyloid pathology (A). A Volcano plot of −log10(p-value) versus log2(fold change) comparing biomarker abundances (NPQ) in samples from A+ participants (n=49) vs. A− controls (n=127). Significant targets are shown in red (higher in A+) or blue (lower in A+) circles. Grey circles represent non-significant targets. B Boxplot distributions of significant NULISAseq targets, separated by A status and visit. P-values on top of the boxplots were for the whole data combining both visits and were determined using linear mixed models (random intercepts) with NPQs as the dependent variable, visit-specific A status as the independent variables, adjusting for covariates age, sex, and APOE ε4 carrier status. Significance determination was based on p-value < 0.005, corresponding to ~8% FDR.

Fig. 3:

Longitudinal association between NULISAseq targets and amyloid pathology (A). A Boxplots illustrating the distribution of yearly biomarker abundance change by A status. P-values were based on two-sided Wilcoxon rank sum tests. B Scatterplots for the correlation between yearly longitudinal Aβ PET SUVR change and baseline biomarker levels. The strength of the correlation was assessed based on Spearman’s ranks. Purple lines indicated the least square regression lines.

Fig. 4:

Association of NULISAseq targets with tau pathology (T). A Boxplots of NULISAseq targets with significant cross-sectional associations with T status, separated by T status and visit. P-values on top of the boxplots were for the whole data combining both visits and were determined using linear mixed models (random intercepts) with NPQs as the dependent variable, visit-specific T status as the independent variables, adjusting for covariates age, sex and APOE ε4 carrier status. Significance determination was based on p-value < 0.005, corresponding to ~9% FDR. B Boxplots illustrating the distribution of yearly biomarker abundance change by T status. P-values were based on two-sided Wilcoxon rank sum tests.

Fig. 5:

Association of NULISAseq targets with neurodegeneration (N). A Heatmaps illustrating the abundance levels of NULISAseq with significant univariate association with N status (unadjusted for covariates). The NPQ values were standardized for each protein target using z-scores. B Boxplots of selected NULISAseq targets, separated by N status and visit. P-values on top of the boxplots were for the whole data combining both visits and were determined using linear mixed models (random intercepts) with NPQs as the dependent variable, visit-specific N status as the independent variables, adjusting for covariates age, sex, and APOE ε4 carrier status. C Boxplots illustrating the distribution of yearly biomarker abundance change by N status. P-values were based on two-sided Wilcoxon rank sum tests.