Head-to-Head Comparison of Aptamer- and Antibody-Based Proteomic Platforms in Human Cerebrospinal Fluid Samples from a Real-World Memory Clinic Cohort

Abstract

High-throughput proteomic platforms are crucial to identify novel Alzheimer's disease (AD) biomarkers and pathways. In this study, we evaluated the reproducibility and reliability of aptamer-based (SomaScan^® 7k) and antibody-based (Olink^® Explore 3k) proteomic platforms in cerebrospinal fluid (CSF) samples from the Ace Alzheimer Center Barcelona real-world cohort. Intra- and inter-platform reproducibility were evaluated through correlations between two independent SomaScan^® assays analyzing the same samples, and between SomaScan^® and Olink^® results. Association analyses were performed between proteomic measures, CSF biological traits, sample demographics, and AD endophenotypes. Our 12-category metric of reproducibility combining correlation analyses identified 2428 highly reproducible SomaScan CSF measures, with over 600 proteins well reproduced on another proteomic platform. The association analyses among AD clinical phenotypes revealed that the significant associations mainly involved reproducible proteins. The validation of reproducibility in these novel proteomics platforms, measured using this scarce biomaterial, is essential for accurate analysis and proper interpretation of innovative results. This classification metric could enhance confidence in multiplexed proteomic platforms and improve the design of future panels.

Keywords: Alzheimer’s disease; Olink; SomaScan; biomarkers; cerebrospinal fluid; mild cognitive impairment; proteomics.

Figure 1

Principal component analysis results. (a) Variance explained by the top 10 PCs. (b) Percentage of cumulative variance explained by the top 300 PCs; the dashed line represents 95% of the explained variance. (c) Pearson correlations of the top 5 PCs with clinical phenotypes according to the nonadjusted model. (d) Linear model associations of the top 5 PCs adjusted by all clinical phenotypes that were included in the model. An asterisk (*) represents a p value lower than 0.05, two asterisks (**) represent a p value lower than 0.01, and three asterisks (***) represent a p value lower than 0.001.

Figure 2

Coefficient of variation (CV) for the SomaScan^® and Olink^® Explore Platforms. (a) Intra- and (b) inter-assay CVs colored by the overlap between the SomaScan^® and Olink^® assays after removing outlier proteins at 1.5-fold IQRs. The Mann-Whitney-Wilcoxon test was applied to explore differences across median values. The red line in the zoom plot represents the median CV for each platform. (c) Percentages of intra- and inter-assay CVs for the SomaScan and Olink Explore platforms.

Figure 3

Distribution of Spearman’s rho values in the correlation analysis, including the sample size of each category (n (%)). (a) Intra-platform correlation between SomaScanA and SomaScanB assays. (b) Inter-platform correlation between SomaScanA and Olink Explore. (c) Inter-platform correlation between SomaScanB and Olink Explore platforms. We established three categories: good (rho > 0.5), moderate (0.5 > rho ≥ 0.3), and poor (rho < 0.3).

Figure 4

Volcano plots of the associations between SomaScan^® proteins and clinical phenotypes. Proteins with intra-assay correlations (rho ≥ 0.5) are colored light blue, and the dashed line represents the significance threshold (p value < 6.860 × 10⁻⁶).

Figure 5

Venn diagram of CSF Aβ42, CSF p-tau, and MMSE associations and enrichment analysis using the WebGestalt tool. (a) Complete set of SomaScan proteins (n = 7289), including 57 overlapping aptamers between the top 500 rankings. (b) Reproducible SomaScan proteins with good intra-correlations (n = 2428), including 88 overlapping aptamers between the top 500 rankings.