Proteomics analysis of extracellular vesicles for biomarkers of autism spectrum disorder

Abstract

Background: Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by symptoms that include social interaction deficits, language difficulties and restricted, repetitive behavior. Early intervention through medication and behavioral therapy can eliminate some ASD-related symptoms and significantly improve the life-quality of the affected individuals. Currently, the diagnosis of ASD is highly limited.

Methods: To investigate the feasibility of early diagnosis of ASD, we tested extracellular vesicles (EVs) proteins obtained from ASD cases. First, plasma EVs were isolated from healthy controls (HCs) and ASD individuals and were analyzed using proximity extension assay (PEA) technology to quantify 1,196 protein expression level. Second, machine learning analysis and bioinformatic approaches were applied to explore how a combination of EV proteins could serve as biomarkers for ASD diagnosis.

Results: No significant differences in the EV morphology and EV size distribution between HCs and ASD were observed, but the EV number was slightly lower in ASD plasma. We identified the top five downregulated proteins in plasma EVs isolated from ASD individuals: WW domain-containing protein 2 (WWP2), Heat shock protein 27 (HSP27), C-type lectin domain family 1 member B (CLEC1B), Cluster of differentiation 40 (CD40), and folate receptor alpha (FRalpha). Machine learning analysis and correlation analysis support the idea that these five EV proteins can be potential biomarkers for ASD.

Conclusion: We identified the top five downregulated proteins in ASD EVs and examined that a combination of EV proteins could serve as biomarkers for ASD diagnosis.

Keywords: ASD; Olink; autism; biomarker; extracellular vesicle.

FIGURE 1

Characterization of plasma EVs isolated from healthy control (HC) and ASD individuals. (A, B) Representative elution profiles of plasma EVs and plasma proteins from HC (A) and ASD individuals (B). EVs are the first to elute, followed by smaller protein complexes. Fractions 2 and 3 were pooled together as EV samples. EV particle numbers and protein concentration in each fraction were determined by NTA and the absorbance at a wavelength of 280 nm, respectively. (C, D) Representative size distribution of plasma EVs determined by NTA. Mean diameter (nm), 121.3 ± 40.32 SD for HC and 120.3 ± 40.53 SD for ASD. (E) EV particle numbers analyzed using NTA; HCs (n = 26) and ASD (n = 81). (F) Plasma protein concentration of HCs (n = 9) and ASD (n = 18). Total plasma proteins eluted in fractions 10 and 11 were measured based on absorbance at 280 nm. (G) Violin plots showing statistical median diameter (X50, nm) of EVs isolated from HC (n = 26) and ASD (n = 81) plasma; 117.6 nm ± 13.99 SD for HC and 126.1 nm ± 13.51 SD for ASD. (H) Morphological characterization of HC and ASD EVs using negative-stain transmission electron microscopy (TEM). Data in (E, F) are means ± SEM. Unpaired two-tailed t-test was used; **, p < 0.01.

FIGURE 2

Differential expression of EV proteins. (A) The volcano plot of the differentially expressed (DE) proteins in plasma EVs isolated from HCs and ASD cases; log₂ fold change (FC) against Limma −log₁₀ BH-adjusted p-value. Color indicates significantly upregulated (red) and downregulated (blue) proteins with FC ≥ 2 and adjusted p-value < 0.05. (B, C) Box plots of the expression level presented as Olink’s normalized protein expression (NPX) for EV proteins (B) and CD63 (C) from control (n = 26) and ASD (n = 60). (B) Only five proteins are significantly downregulated with FC ≥ 2 and adjusted p-value < 0.05. (D) Gene Ontology (GO) enrichment analysis for cellular components of downregulated proteins in ASD EVs. The GO cut-off criteria included adjusted p-value < 0.05 and FC ≥ 1.5. (E) GO enrichment analysis for biological process of dysregulated proteins in ASD EVs. The GO cut-off criteria included q (adjusted p value) < 0.05 and 1.5 FC.

FIGURE 3

Machine learning outcome. (A) A Venn diagram of the overlapping proteins selected by MUVR, VSURF, and Boruta methods. (B) An ROC curve of the true positive rate versus the false positive rate for different threshold values of the classifier. (C) A table summarizing the performance metrics of the best classifier, which is the random forest using the intersect of the features selected by MUVR, Boruta and VSURF.