The Liquid Biopsy Frontier Over 70% of lung cancer patients are diagnosed at advanced stages when curative surgery is no longer possible. Liquid biopsy - cancer detection from blood samples - offers the potential for earlier, non-invasive diagnosis, but current approaches using circulating tumor DNA or serum protein markers lack sufficient sensitivity and specificity for widespread screening.
Extracellular Vesicles as Biomarker Carriers Extracellular vesicles (EVs) are tiny membrane-enclosed particles shed by cells into the bloodstream. Tumor cells release EVs that carry proteins, nucleic acids, and lipids reflecting the molecular characteristics of the parent cancer cell. These tumor-derived EVs (TDEVs) offer rich, stable biomarker content that survives circulation and can be captured from blood.
Study Approach Yuan and colleagues at the Chinese Academy of Sciences used CD155 - a protein overexpressed on NSCLC cell surfaces - to specifically enrich TDEVs from patient plasma, then applied comprehensive proteomics and machine learning to identify a diagnostic protein biomarker panel capable of distinguishing NSCLC patients from healthy individuals.
CD155 as a Bait Protein CD155 (also called PVR - poliovirus receptor) is a transmembrane glycoprotein that is strongly overexpressed in NSCLC tumors and their shed EVs but nearly absent from healthy tissue EVs. By using anti-CD155 antibodies conjugated to magnetic beads, the researchers could selectively capture CD155-positive TDEVs while leaving normal cell-derived EVs in solution.
Clinical Samples The study used plasma samples from 355 donors: 108 patients with LUAD, 31 with LUSC, and 216 healthy individuals from Shiyan Renmin Hospital. Samples were processed through standardized EV isolation procedures including sequential ultracentrifugation, with quality control at each step to ensure sample integrity.
Discovery and Verification Cohorts The study was designed in two phases: a discovery cohort using pooled plasma samples for untargeted DIA mass spectrometry proteomics, and a verification cohort using individual patient samples analyzed by targeted PRM (parallel reaction monitoring) mass spectrometry - a more precise, quantitative approach suitable for validation.
Proteomics Scale Using data-independent acquisition (DIA) mass spectrometry on the Orbitrap Eclipse instrument, the researchers identified over 2,000 proteins in each group. Comparison between NSCLC TDEVs and healthy control EVs revealed 281 differentially expressed proteins (DEPs) - 126 upregulated and 155 downregulated in the NSCLC group.
CD155 Specificity Confirmed CD155 itself was consistently detected in CD155-immunocaptured TDEVs but was absent or extremely low in CD63-immunocaptured EVs from healthy individuals. This confirmed that CD155 immunocapture successfully enriched genuinely tumor-derived vesicles rather than non-specifically capturing all plasma EVs.
Machine Learning Ranking Random forest analysis was applied to the 281 DEPs to rank proteins by their importance for classifying NSCLC versus healthy samples. This identified a priority list of candidates for targeted verification, reducing the discovery-to-validation funnel from 281 to 50 top candidates.
Three-Algorithm Approach For biomarker selection from the 49 verified candidate proteins, three complementary machine learning algorithms were applied: LASSO regression for feature selection (identifying the sparsest informative subset), random forest for feature importance ranking (identifying the most discriminative proteins), and Boruta (a wrapper method identifying all statistically significant features). Using all three algorithms provides more robust biomarker identification than any single method.
Panel Optimization Each algorithm produced a different candidate list. The researchers then systematically tested combinations from all three algorithms using confusion matrix analysis to find the combination with the best sensitivity and specificity balance. Proteins that did not contribute additional diagnostic value (IGLL1, CP, MYCBP2, PRPS2) were excluded.
Final Panel Performance The optimized 7-protein panel - MVP, GYS1, SERPINA3, HECTD3, SERPING1, TPM4, and APOD - achieved AUC of 1.0 with 100% sensitivity and specificity in ROC curve analysis using the targeted proteomics data, and 92.3% sensitivity and 88.9% specificity in confusion matrix analysis. Western blotting independently confirmed the expression trends for all 7 proteins.
Oncogenic Proteins MVP (Major Vault Protein) is upregulated in NSCLC and associated with chemotherapy resistance. GYS1 (Glycogen Synthase 1) drives metabolic reprogramming in cancer cells, promotes tumor-associated macrophage M2 polarization, and correlates with poor prognosis. SERPINA3 promotes tumor progression through inflammation-mediated angiogenesis and immune suppression.
Cancer Progression Factors HECTD3 is an E3 ubiquitin ligase overexpressed in multiple cancers that promotes tumor growth by polyubiquitinating caspases and c-Myc. SERPING1 is a complement regulator whose high expression in NSCLC brain metastases correlates with poor prognosis. TPM4 promotes lung cancer cell migration through F-actin remodeling.
Nuanced Biomarker APOD (Apolipoprotein D) shows context-dependent cancer roles - downregulated in some cancers (liver, colon) but upregulated in gastric cancer and NSCLC. Inconsistent APOD results between discovery and verification cohorts highlight the need for further validation in larger cohorts with multiple detection methods.
Diagnostic Potential The near-perfect ROC performance and 92.3% sensitivity in a real patient cohort suggest this panel has genuine potential as a blood-based NSCLC diagnostic. The EV-based approach could complement existing CT screening - potentially identifying cancers too small to appear on CT, or clarifying the malignancy risk of detected nodules.
Limitations to Address The study population was drawn from a single Chinese hospital, limiting demographic diversity. The discovery phase used pooled samples (which reduces individual variability), and the verification cohort size (71 individual samples) is modest. Larger, multi-center validation with diverse populations is essential before clinical translation.
Future Development Future work should develop the TDEV-based assay into a standardized clinical test format, incorporate the panel into multiplex point-of-care platforms, and explore whether dynamic changes in the 7-protein panel during treatment predict response or resistance - potentially enabling ongoing treatment monitoring from blood rather than requiring serial CT scans.