Tissue‐derived extracellular vesicle profiling identifies GLUT1 enabling ultrasensitive circulating quantification and early detection of non‐small cell lung cancer

Dear Editor, Extracellular vesicle (EV)-based liquid biopsy is promising for early detection of non-small cell lung cancer (NSCLC), but hindered by the low abundance and limited cancer specificity of circulating EV cargoes. Using multi-omics integration, we identified glucose transporter 1 (GLUT1) as a membrane-localised biomarker in tissue-derived EVs and achieved ultrasensitive circulating quantification with plasma samples, potentially discriminating NSCLC from benign lesions, including in computed tomography (CT)-suspected indeterminate pulmonary nodules (IPNs). EVs from body fluids or cell lines have been well studied, whereas tissue-derived EVs may offer greater cancer specificity by concentrating tumour-proximal secretomes and minimising systemic contamination, thereby providing more representative biological insights.1, 2 In our study, tissue-derived EVs were isolated by ultracentrifugation with density gradient flotation and subjected to proteomic profiling (Figure 1). Tissue-derived EVs exhibited comparable characteristics to cell line-derived EVs, and demonstrated distinct protein profiles between cancer and non-cancer groups (Figure 2A–D). A total of 904 differentially expressed proteins (DEPs) were initially screened (NUH group; Figure 2E–G). Enrichment analyses linked the top-ranked DEPs to microenvironment-related processes and glycolysis/gluconeogenesis, facilitating subsequent biologically grounded biomarker selection (Figure 2H and I). To prioritise more cancer-enriched candidates from tissue-derived EVs, we integrated The Cancer Genome Atlas (TCGA) transcriptomes and identified 14 431 differentially expressed genes (DEGs) (TCGA group; Figure S1A–D). Moreover, we curated a reference set of 2074 genes encoding membrane-localised proteins from the Human Protein Atlas (HPA) (HPA group). Intersecting the NUH, TCGA, and HPA groups identified 24 DEGs, among which 18 exhibited consistent dysregulation patterns across proteomic and transcriptomic levels, and were enriched in glucose transport, metabolism- and hypoxia-associated pathways (Figures 2J and S1E and F). Analysing 106 matched cancer and non-cancer tissue pairs from TCGA group, eight genes with cancer-specific dysregulation were selected (Figure 3A), with GLUT1 (encoded by SLC2A1) exhibiting the highest importance score (Figure 3B), the highest expression levels (Figures 3C and S2A), the strongest diagnostic performance (Figures 3D and S2B) and prognostic significance (Figure 3E). Immunohistochemistry in this study also supported GLUT1 enrichment in lung malignancies/lesions (Figure S3A). GLUT1 is overexpressed across diverse malignancies, where its upregulation enhances glucose uptake and drives aerobic glycolysis.3, 4 Notably, accumulating evidence indicates that GLUT1-related EVs contribute to metabolic reprogramming and oncogenesis by transferring glycolytic phenotypes to recipient cells, providing a biological rationale for exploring EV-derived GLUT1 as a biomarker.5, 6 Through immunofluorescence (HCC827 and A549 cells) and immunogold labelling transmission electron microscopy (cell-derived EVs), we observed membrane-localised signals of GLUT1 on both cell and vesicle surfaces (Figure 3F and G). To overcome the challenge of detecting scarce signals, we applied an ultrasensitive digital EV assay (UDA) incorporating tyramide signal amplification, enabling rapid and sensitive detection of low-abundance EV-derived proteins from minimally processed plasma (Figure 1 and Tables S1 and S2).7 In the pilot cohort (38 histologically confirmed NSCLC patients and 9 non-cancer controls), the UDA results demonstrated that CD63+ EV-derived GLUT1 formed a distinct clustering pattern from CD9+ and CD81+ counterparts (Figure 3H and Table S3). Moreover, CD63+ EV-derived GLUT1 displayed significantly different signals and superior discrimination between cancer and non-cancer samples, including in stage I patients (Figures 3I and S3B). As reported, EVs bearing CD63 along with one or both of the other tetraspanins probably represent endosome-derived exosomes, which are thought to play prominent roles in cargo transport and intercellular communication.8, 9 We also established an overall cohort of 121 NSCLC patients and 36 non-cancer controls (Table S3). The real-world overall cohort included IPNs that were radiologically suspicious but could not be definitively classified as malignant or benign using CT alone, reflecting the commonly encountered uncertainty in clinical diagnostic practice. The UDA results revealed that GLUT1 levels were not correlated with sample batch (r = .25), and were significantly elevated in cancer samples, including in stage I patients (p < .001) (Figure 4A–C). No significant differences in GLUT1 levels were observed across histological subtypes or pathological stages (Figure S3C–E). An exploratory set of inflammatory diseases also indicated lower GLUT1 levels than the cancer group (Figure S3F). The optimal predictor set was selected by LASSO logistic regression with fivefold cross-validation. For parsimony and clinical interpretability, three variables (GLUT1, CEA, and CYFRA 21-1) were retained for the final model, which demonstrated diagnostic performance comparable to the full six-variable model (Figures 4D–H and S4A and B). Specifically, in all patients, CD63+ EV-derived GLUT1 achieved an AUC of .804 in ROC analysis, outperforming CEA (AUC = .633) and CYFRA 21-1 (AUC = .710) in discriminating cancer from non-cancer samples (Figure 4G). Diagnostic performance was improved by the three-marker panel (AUC = .877; Optimism-corrected AUC = .863) (Figure 4G). Notably, even in stage I patients, CD63+ EV-derived GLUT1 maintained superior performance (AUC = .822), while the three-marker panel yielded the improved discrimination (AUC = .861; Optimism-corrected AUC = .841) (Figure 4H). Favourable calibration curves and low Brier scores ( .113 and .142) indicated satisfactory overall calibration and accuracy (Figure 4I and J and Table S4). Decision curves showed higher net benefit of the three-marker panel than treat-all or treat-none strategies (Figure 4K and L). Further subgroup analysis of stage IA patients (tumour size < 3 cm) showed consistent findings. CD63+ EV-derived GLUT1 achieved an AUC of .850 while the three-marker panel further improved discrimination (AUC = .879; Optimism-corrected AUC = .855) with acceptable calibration, Brier score, and net benefit (Figure S4C–E and Table S4). Distinguishing malignant from benign nodules remains a major unmet need in thoracic oncology, often leading to unnecessary invasive procedures or delayed diagnosis.10 By complementing imaging with molecular information, circulating CD63+ EV-derived GLUT1 may potentially contribute to preoperative malignancy risk assessment and support more informed clinical decision-making in this diagnostically challenging population. In conclusion, our findings support a biomarker discovery workflow integrating proteomic profiling of tissue-derived EVs with ultrasensitive digital qualification in circulating EVs (Figure 1). We identified CD63+ EV-derived GLUT1 as a circulating biomarker for NSCLC, including stage I disease, with promising diagnostic performance that was further enhanced when combined with conventional serum biomarkers. This proof-of-concept study indicates the feasibility of translating tissue-derived EV proteomic signals into non-invasive circulating biomarkers; however, external validation in larger cohorts and mechanism investigations are critically warranted in subsequent work. Conception and design, H.H. and T.K. Data acquisition, H.H., Y.I., Y.N., H.W., H.T., T.R., Y.K., K.N., Y.K., H.U., Y.A., M.K. and E. A-I. Data analysis and interpretation, H.H., T.K., S.N., T.M., A.T., A.Y. and T. C-Y. Supervision, T.K. and T. C-Y. Funding acquisition, T.K. and T. C-Y. Writing-original draft, H.H. and T.K. Project administration, T.K. and T. C-Y. Writing-review and editing, T.K. S.N., T.M., A.T., A.Y. and T. C-Y. Final approval of manuscript: all authors. The authors wish to acknowledge the Division for Medical Research Engineering, Nagoya University Graduate School of Medicine for the usage of NanoSight NS300 system, Optima XE-100, and BD FACS Canto™ II Flow Cytometer, and we also thank Dr. Itakura (Technical Center, Nagoya University) and Dr. Lu (Department of Pathology, Nagoya University Graduate School of Medicine) for technical support and data acquisition. In addition, we gratefully acknowledge Enago company (www.enago.jp) for English language editing. The authors declare that they have no competing interests. This work was supported by JSPS KAKENHI (Grant number: 22K16566 and 24K12009), and the Hori Sciences and Arts Foundation and Nagoya University Hospital Funding for Clinical Research. The current study was conducted in accordance with the ethical guidelines and received approval from the Institutional Review Board of Nagoya University Hospital (No.2022-0107). The data that support the findings of this study are available from the corresponding authors upon reasonable request. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.