Identification of a Biomarker Panel in Extracellular Vesicles Derived From Non-Small Cell Lung Cancer (NSCLC) Through Proteomic Analysis and Machine Learning

Abstract

Antigen fingerprint profiling of tumour-derived extracellular vesicles (TDEVs) in the body fluids is a promising strategy for identifying tumour biomarkers. In this study, proteomic and immunological assays reveal significantly higher CD155 levels in plasma extracellular vesicles (EVs) from patients with non-small cell lung cancer (NSCLC) than from healthy individuals. Utilizing CD155 as a bait protein on the EV membrane, CD155+ TDEVs are enriched from NSCLC patient plasma EVs. In the discovery cohort, 281 differentially expressed proteins are identified in TDEVs of the NSCLC group compared with the healthy control group. In the verification cohort, 49 candidate biomarkers are detected using targeted proteomic analysis. Of these, a biomarker panel of seven frequently and stably detected proteins-MVP, GYS1, SERPINA3, HECTD3, SERPING1, TPM4, and APOD-demonstrates good diagnostic performance, achieving an area under the curve (AUC) of 1.0 with 100% sensitivity and specificity in receiver operating characteristic (ROC) curve analysis, and 92.3% sensitivity and 88.9% specificity in confusion matrix analysis. Western blotting results confirm upregulation trends for MVP, GYS1, SERPINA3, HECTD3, SERPING1 and APOD, and TPM4 is downregulated in EVs of NSCLC patients compared with healthy individuals. These findings highlight the potential of this biomarker panel for the clinical diagnosis of NSCLC.

Keywords: NSCLC; biomarkers; diagnosis; extracellular vesicles; proteomics.

FIGURE 1

Characterization of EVs derived from cell lines and clinical plasma. (A) Representative TEM images of purified EVs. Scaled bars: 100 nm. Left: EVs derived from cell culture supernatant. Right: EVs isolated from plasma. (B) Analysis of the size distribution and concentrations of isolated EVs using nanoparticle tracking analysis. Left: EVs isolated from cell culture supernatant. Right: EVs isolated from plasma. (C) EV biomarker detection using western blotting.

FIGURE 2

Proteomic profiling of EVs isolated from NSCLC and control cell lines using DIA mass spectrometry. (A) PCA of the protein profiles of NSCLC‐EVs and control‐EVs. (B) Venn diagram showing the identified proteins in NSCLC‐EVs and control‐EVs. (C) Venn diagram showing the identified EV proteins in each sample of NSCLC and control groups. (D) Volcano plots representing the differences in protein expression between NSCLC‐EVs and control‐EVs. (E) Heatmap showing hierarchical clustering of the top 30 proteins with the most significant differences in expression between EVs from NSCLC and control cell lines. The colour scale represents normalized protein expression levels. (F) Heatmap displaying the 15 most abundant proteins exclusively detected in NSCLC‐EVs (top) and control‐EVs (bottom). The colour scale represents the relative abundance of these proteins, white indicating proteins that were not detected.

FIGURE 3

Elevated CD155 expression in EVs from NSCLC cells and patient plasma compared with healthy individuals. (A) Western blotting analysis of CD155 in EVs isolated from NSCLC and control cell lines. (B) Flow cytometry analysis of CD155 and CD63 in EVs isolated from H1975 and BEAS‐2B cells. (C) Western blotting analysis of CD155 abundance in EVs from the plasma of patients with NSCLC and healthy individuals in three independent tests. (D) Quantitative assessment of the fold change in CD155 expression from panel C, normalized to whole protein. (E) CD155 expression in EVs isolated from plasma of patients with NSCLC (LUAD, n = 20; LUSC, n = 4) and healthy individuals (n = 16) using ELISA. (F) ROC curve analysis of the diagnostic potential of EV‐CD155 in distinguishing patients with NSCLC from healthy individuals. (G) Pearson correlation analysis of CD155 expression in EVs and tumour size in patients with NSCLC (n = 24). (H) IHC analysis for CD155 expression in NSCLC tissue microarrays. Left: IHC staining images of paired tumour tissues (TT) and adjacent tumour tissues (ATT) (n = 35). Right: quantitative analysis of IHC intensity for CD155 (n = 49). T1, T2, and T3 in the quantification data refer to the tumour stages defined by the TNM classification system of NSCLC.

FIGURE 4

Cellular origin and membrane localization of CD155 in EVs. (A) Immunofluorescence analysis of H1975 cells showing CD155 displayed higher co‐localization with CD63 and Rab5 than Rab7. Nuclei were stained with DAPI. Scale bars: 10 µm. (B) Quantitative analysis of Pearson's correlation coefficients demonstrating the co‐localization between CD155 with CD63, Rab5, and Rab7 in H1975 cells. (C) Western blotting analysis of cellular and EV CD155 treated with different endoglycosidases. PNGase F effectively removes the oligosaccharide chains of CD155, whereas these chains are resistant to Endo H cleavage. (D) Representative iEM images of EVs isolated from H1975 and BEAS‐2B cells. EVs were stained with anti‐CD155 antibodies. Scale bars: 100 nm. (E) Western blotting analysis of the immunocaptured EVs derived from H1975 cells using antibody against CD63 or CD155. (F) Representative iEM images of plasma EVs demonstrating CD155 localized on the EV membrane rather than within the lumen. Scale bars: 100 nm.

FIGURE 5

DIA proteomic profiling of immunocaptured EVs from plasma of patients with NSCLC and healthy control groups. (A) Western blotting analysis of the immunocaptured EVs from the total plasma EVs from patients with NSCLC and healthy control groups. (B) Venn diagram showing the identified proteins in immunocaptured EVs from the two groups. (C) PCA results of the identified proteins in immunocaptured EVs from NSCLC and healthy control groups. (D) Volcano plots demonstrating the significant differences in protein expression of immunocaptured EVs from NSCLC and healthy control groups. (E) The top 50 important proteins identified using the random forest algorithm.

FIGURE 6

Identification and evaluation of the candidate biomarkers in EVs for NSCLC screening. (A) Violin plots showing the abundance of the 21 DEPs in EVs between NSCLC patients and healthy individuals. (B) Feature selection from the 21 DEPs using LASSO regression. (C) Random forest algorithm evaluating the importance of the 21 DEPs in the classification of NSCLC patients and healthy individuals. (D) Boruta algorithm evaluating the importance of the 21 DEPs in the classification of NSCLC patients and healthy individuals. (E) Confusion matrix showing the classification of the two groups with the selected 7‐ and 21‐protein panels. (F) Western blotting analysis of the expression of seven biomarkers in the plasma EVs (LUAD, n = 16; LUSC, n = 4; healthy individuals, n = 22). (G) Statistical analysis of the relative expressions of the seven biomarkers from panel G, normalized to whole protein. Dashed lines, black dots, and grey dots represent p = 0.05, p < 0.05 and p > 0.05, respectively. (H) ROC curve analysis of the panel of seven proteins in distinguishing between NSCLC patients and healthy individuals. (I) The classification of NSCLC patients and healthy individuals based on the panel of 7 and the panel of 21 DEPs using PCA.

FIGURE 6

Identification and evaluation of the candidate biomarkers in EVs for NSCLC screening. (A) Violin plots showing the abundance of the 21 DEPs in EVs between NSCLC patients and healthy individuals. (B) Feature selection from the 21 DEPs using LASSO regression. (C) Random forest algorithm evaluating the importance of the 21 DEPs in the classification of NSCLC patients and healthy individuals. (D) Boruta algorithm evaluating the importance of the 21 DEPs in the classification of NSCLC patients and healthy individuals. (E) Confusion matrix showing the classification of the two groups with the selected 7‐ and 21‐protein panels. (F) Western blotting analysis of the expression of seven biomarkers in the plasma EVs (LUAD, n = 16; LUSC, n = 4; healthy individuals, n = 22). (G) Statistical analysis of the relative expressions of the seven biomarkers from panel G, normalized to whole protein. Dashed lines, black dots, and grey dots represent p = 0.05, p < 0.05 and p > 0.05, respectively. (H) ROC curve analysis of the panel of seven proteins in distinguishing between NSCLC patients and healthy individuals. (I) The classification of NSCLC patients and healthy individuals based on the panel of 7 and the panel of 21 DEPs using PCA.