Non-small cell lung cancer (NSCLC) is the leading cause of cancer deaths worldwide, and adenocarcinoma is the most common cancer type (1). Surgery is the cornerstone of treatment of early-stage and resectable locally advanced NSCLC (2). Nevertheless, patients frequently develop local and/or distant relapses despite complete surgical resection. Rate of recurrence is approximately 25% for stage I-II NSCLC (3) and 50–70% for stage III NSCLC, treated with multimodality therapy (3, 4).

Thus, identification of prognostic markers of early relapse after surgery is a major challenge. Currently used imaging techniques are not sensitive enough to detect early-stage metastasis. Several liquid biopsy biomarkers have been studied to predict resected cancer recurrence, including circulating tumor cell (CTC) (5–7), circulating tumor DNA (ctDNA) (8, 9) but none are currently used in practice for this purpose. Among them, extracellular vesicles (EVs) are nanoscale lipid bilayer vesicles secreted by all cells, acting as key mediators of intercellular communication (10). EVs released from tumor cells, including exosomes, are involved in premetastatic niche establishment, tumor progression, angiogenesis, and immune response modulation (10, 11). Indeed, they are known for their capacity to transport many cellular components, such as mRNA, miRNA, lipids and proteins. Thus, EVs cargo could reflect the characteristics of the cell of origin.

Due to their abundance and stability in biofluids, EVs analysis could be more relevant over conventional liquid biopsy (10–12). Nevertheless, due to their relative abundance in biofluids compared to CTCs or ctDNA, EVs offer a promising alternative approach for biomarker analysis in NSCLC. It's important to note that tumor-derived EVs constitute only a small fraction of circulating EVs in the body (13, 14), particularly in non-metastatic stages of the disease. To address this limitation, numerous studies have shown that tumor-draining vein (TDV) samples are enriched with various oncological biomarkers compared to peripheral blood samples. This difference may be attributed to the dilution of tumor-derived EVs in systemic blood and their degradation by the immune system. (15).

Navarro and colleagues were the only research team to investigate EVs from the tumor-draining pulmonary vein in patients undergoing surgery for NSCLC (16–18). Their findings revealed that patients with smaller pulmonary vein exosome had a higher risk of post- surgery relapse and poorer overall survival (16). Furthermore, miRNA analysis of EVs isolated from the pulmonary vein in the operated patient cohort enabled to identify potential biomarkers for relapse (17, 18).

To date, proteomic analysis of EVs from vascular bed of lung cancer has not been reported previously in the literature. However, studies on the protein cargo of peripheral blood EVs in lung cancer patients have revealed proteomic signatures associated with both diagnosis (19–22) and prognosis (23, 24).

Based on these research approaches, we hypothesized that blood samples from the pulmonary vein are enriched with primary tumor-derived EVs. Consequently, we conducted the first proteomic analysis of EVs from the tumor-draining vein of operated lung adenocarcinoma-patients, comparing the protein cargo with that of peripheral blood EVs and cancer tissue.

Following resection both tumor and non-tumor tissues were snap-frozen in liquid nitrogen and stored at -80°C. Prior to surgery, a nurse anesthesiologist collected 20 mL of blood from a peripheral vein using EDTA tubes. Before initiating the resection, the thoracic surgeon collected 20 mL of blood from the pulmonary tumor-draining vein (upper pulmonary vein for upper and middle lobe tumors, lower pulmonary vein for lower lobe tumors) using EDTA tubes. Plasma was then separated by centrifugation and stored at -80°C.

EVs were purified according to the standard protocol described by Théry et al., (26). Briefly, 0.5 mL of plasma was diluted with an equal volume of phosphate-buffered saline (PBS) and centrifuged at 2000g for 20 minutes (min). The resulting supernatant was then centrifuged at 16 500g for 45 min. Subsequently, the supernatant underwent ultracentrifugation at 120 000g for 2 hours (h) using an Optima MAX-XP Beckman coulter ultracentrifuge with a MLA-130 rotor. The pellet was then resuspended in PBS and subjected to ultracentrifugation at 120 000g for 2 h. Finally, the pellet was resuspended in PBS for nanoparticle tracking analysis (NTA) or flow cytometry analysis.

For proteomic analysis, EVs were isolated from plasma using size-exclusion chromatography (SEC). IZON qEV2 Gen2 columns (35 nm series) were used according to the manufacturer’s instructions. Briefly, 2 mL of plasma were loaded onto the column for each patient and 5 fractions of 2 mL each were eluted in PBS. The eluted plasma-EVs fractions were then ultracentrifuged at 120 000g for 2 h. The resulting EV-containing pellets were resuspended in 20 µL of cell lysis buffer from Cell Signaling Technology and stored at -80°C.

Liquid Chromatography tandem MS (LC-MS/MS) analysis identified a total of 3878 proteins with a Peptide Spectrum Match (PSM) strictly superior to 3. Among these, 2758 proteins were quantified in all patient samples (19 tumor and 20 paired non-tumor tissues) while 3553 proteins were detected in at least 70% of patients. Principal component analysis (PCA) analysis revealed a clear segregation between tumor (T) and non-tumor (NT) tissue proteomes, suggesting that tumor progression significantly alters protein expression profiles (Fig. 1A). Among the 3553 proteins, 148 proteins (31.75%) were significantly increased while 318 proteins (68.25%) were decreased in tumors compared to non-tumor tissues, with at least two-fold change (Additional Table 1). Among these, 72 proteins showed high differential expression (> 4-fold increase or decrease) (Fig. 1B). Top-10 of most upregulated in T versus NT tissues were PYCR1, GOLM1, TUBB3, S100B, TMEM165, OCIAD2, CYP1B1, CPD, UGT1A6 and PES1 (Fig. 1B, Additional Table 1). To further explore tumor-driven changes in protein expression, clustering analysis was performed on tumor and non-tumor tissue proteomes, revealing two distinct clusters (Fig. 1C): cluster 1 includes 95% of non-tumor tissues and 3 early stage tumor tissues; cluster 2 includes 84% of tumor tissues and 1 non-tumor tissue. Cluster 1 can be subdivided in two subgroups: cluster 1a and cluster 1b. Cluster 1a displays completely opposite expression patterns compared to cluster 2. Cluster 1b is a mixed intermediate subgroup, with both tumor and non-tumor samples, showing an intermediate expression profile, especially for protein identified as downregulated in tumor versus non-tumor tissues. This finding reinforces the profound proteomic shift between tumor and non-tumor tissues, as previously demonstrated by PCA analysis (Fig. 1A). Functional enrichment performed on 148 up-regulated proteins in lung adenocarcinoma (LUAD) compared to non-tumor tissues highlights a modulation of many metabolic processes, as glycolytic and aminoacid metabolism pathway or biosynthesis of proteins, a typical feature of cancer (Additional Fig. 1A). So, the same analysis conducted on 318 downregulated proteins in LUAD, put in evidence functions associated to cell adhesion, reorganization of cytoskeleton and cell migration. Interestingly, enrichment performed on GO Cellular Component (GO-CC) from up- and downregulated proteins highlights that extracellular proteins associated to exosome are significantly highest (Additional Fig. 1A).

In order to confirm our data obtained on a moderate number of patients (n = 20), we extracted proteomic data of 112 lung adenocarcinoma and 101 healthy lung tissues (https://pdc.cancer.gov/pdc/study/PDC000153) from the National Cancer Institute Proteomic Data Commons (NCI-PDC) (35). The hierarchical clustering based on highlighted protein levels in our study allows discriminating tumor and healthy samples groups (Additional Fig. 2B and 2D). By comparing signal intensities obtained from increased (Additional Fig. 2A and 2B) and decreased (Additional Fig. 2C and 2D) proteins, we observed that the great majority of deregulated proteins characterized in our study are also identified from NCI data. Among 444 proteins found in NCI-PDC000153, 15% (66 proteins) showed a similar level between LUAD and normal tissue when the absolute fold change was fixed to 1.2. In our study, among these 66 proteins, 36 were upregulated and 30 downregulated proteins (Additional Table 1).

These results suggest that proteome change in tumor tissue could have an impact on the content of extracellular vesicles produced by tumor cells. To further explore the significance of these findings, we have compared in a second time the tumor-derived proteome clusters with proteins identified in EVs from tumor-draining and peripheral veins to assess their potential roles in lung cancer progression and tumor cell-derived EV cargo.

The schematic workflow outlining the different stages of this study, from blood sample collection during tumor resection to the in silico analysis of the EV-derived proteome, is presented in Fig. 2. Following purification, EVs isolated from the tumor-draining vein (TDV) or peripheral vein (PV) of lung cancer patients were characterized based on their size distribution using nanoparticle tracking analysis (NTA). As shown in Fig. 3A, EVs from both TDV and PV samples predominantly ranged between 50 and 150 nm in diameter, consistent with the typical size of small EVs. Additionally, flow cytometry analysis (Fig. 3B) confirmed the presence of CD81 tetraspanin protein, commonly used as an EVs marker. These findings validate the successful purification and characterization of EVs from both TDV and PV plasma samples.

NTA analysis revealed a significantly higher EV concentration in TDV plasma samples compared to PV plasma samples (17.0x10⁹ vs 4.4x10⁹ particles/mL respectively) (Fig. 4A). Additionally, EVs in TDV plasma were smaller than those in PV plasma with a mean size of 98.7+/-22.9 nm vs 120.6+/-23.4 nm and a mode size of 74.7+/-17.5 vs 97.2+/-25.2 nm respectively (Fig. 4A). Interestingly, further analysis categorized EVs into three size subpopulations: inferior to 100 nm, comprised between 100 to 150 nm and superior to 150 nm. TDV plasma samples showed a significant enrichment of EVs across all three size categories compared to PV plasma samples. Furthermore, EVs with a size inferior to 100 nm were significantly enriched in TDV plasma samples compared to the 100–150 nm EV subpopulation (p < 0.01) (Fig. 4B). Although EVs inferior to 100 nm in TDV plasma tended to be more abundant than EVs superior to 150 nm, this difference was not significant.

EVs concentration and size were not correlated with tumor size or other clinical characteristics. Within the cohort, four (20%) patients diagnosed with pIIIA stage relapsed after surgery and three of them developed distant metastases. No death was reported during follow-up. Risk factors for cancer recurrence included: operative blood loss (median 105 mL vs 40 mL, p = 0.047), tumor size (72 +/- 21 mm vs 31+/- 14 mm, p = 0.024) and pN + status (p = 0.013).

Whatever the recurrence status of patients, no significant difference was observed in EV concentration, median size, or mode size in PV plasma samples (p > 0.19 for all comparisons). Similarly, these EVs characteristics in TDV samples were not different between recurrent and no recurrent patients (respectively, 18.0 vs 16.7, p = 0.78; 96.8 vs 99.2, p = 0.84; 72.0 vs 75.4, p = 0.89).

Differential protein expression in EVs from tumor draining-vein versus peripheral vein.

Liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis identified 1,890 proteins in EV cargo with a Peptide Spectrum Match (PSM) strictly superior to 3. Among these, 761 proteins were quantified across all patient samples (19 TDV-EV samples and 14 PV-EV samples) while 1410 proteins were detected in at least 70% of patients. In addition, 87.3% of the EV-associated proteins detected in our study have been previously reported in vesicle-related databases: 78% were found in VesiclePedia and Exocarta databases, 9.3% were retrieved in at least one of these 2 databases. Interestingly, 12.7% of proteins (179) were newly identified in extracellular vesicles (Additional Fig. 3). The proteomic profile of EVs from TDV appeared distinct from that of EVs from PV, as demonstrated by principal component analysis (PCA), which showed a clear separation between the two groups (Fig. 5A). Among the 1410 proteins analyzed, 176 proteins (12.5%) were significantly increased and 193 proteins (13.7%) were decreased in TDV-EVs compared to PV-EVs, with at least two-fold change (Additional Table 2). Among these, 94 proteins showed high differential expression (> 4-fold increase or decrease). Top-10 of most upregulated in TDV-EVs versus PV-EVs were MUC5B, CCT6A, UNC45A, IGHV5-51, ATP1A1, STXBP3, SSC5D, VDAC1, F12, SRPRB (Fig. 5B, Additional Table 2). Clustering analysis of protein expression in EVs revealed 2 distinct clusters: cluster 1 included a majority of PV-derived EVs (13 versus 4 EVs from TDV) and cluster 2 composed almost exclusively of TDV EVs except for one PV-EV sample (Fig. 5C). This analysis confirmed the presence of a specific proteomic signature in EVs from TDV compared to PV, identifying clusters of overexpressed proteins unique to TDV-EVs.

The analyses of Gene Ontology (GO) annotation, KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways and Reactome Gene Sets indicated that the majority of proteins in TDV-derived EVs were associated with extracellular exosomes, suggesting that identified proteins belong to exosome cargo (Fig. 6A). The top 5 enriched terms in biological process (BP) are involved in protein transport, regulation of the immune system and actin cytoskeleton remodeling. Enriched terms associated to molecular function are closely linked to enzyme regulation and binding or binding to carbohydrate derivative. Regarding KEGG pathway, at least 50 proteins were implicated in the Rho GTPase signaling pathway which plays a role in mechano-signaling and might regulate osimertinib resistance and brain metastasis in lung cancer (36).

The significance of these functional clusters was confirmed in the network showing protein-protein interaction (Fig. 6B). A densely connected protein-protein interaction (PPI) network was generated, comprising 2746 edges (interactions) between 342 proteins (nodes). The density of this network suggests that all proteins are closely and functionally linked supporting the functional clustering observed in GO analyses. Using CytoHubba, a Cytoscape plugin, we identified the top hub proteins based on their degree of connectivity (Fig. 6C): 9 upregulated proteins in TDV-EVs (FN1, KNG1, GRB2, HSPD1, VDAC1, CLTC, CANX, DBI, UQCRC2) and 13 downregulated proteins (ACTB, ENO1, SRC, HSPA8, ITGB2, LYN, ATIC, PRKACB, HSPA5, YWHAZ, CFL1, SOD1, PKM).

Comparison of proteome profiles between tumor-draining vein (TDV)-derived EVs and lung tumor tissues

To identify common deregulated proteins between TDV-derived EVs and lung tumor tissues, we intersected the differentially expressed proteins (DEP) across 4 conditions [upregulated (148) and downregulated (318) proteins in tumor versus non-tumor tissues with upregulated (176) and downregulated (193) proteins in TDV- versus PV-derived EVs] (Fig. 7A). This intersection revealed 26 proteins that were commonly deregulated in both TDV-EVs and tumors. Only one protein, SRPRB (Signal recognition particle receptor subunit beta), was commonly upregulated in both TDV-EVs and lung tumor tissues. In contrast, 12 proteins (AHNAK, APOA2, APOB, APOC1, COL6A1, CSRP1, HLA-E, ITIH2, MCEMP1, PCYOX1, PGLYRP2, SERPING1) were found at the intersection of upregulated proteins in TDV-EVs and downregulated proteins in lung tumor tissues (Fig. 7A). A striking observation was that 46% of deregulated proteins common to EVs and tumors were downregulated in tumors but upregulated in TDV-EVs. This suggests that tumor cells may actively load these proteins into EVs as a mechanism to eliminate tumor-suppressive factors that could otherwise inhibit tumor growth and progression. To further investigate the abundance of tumor-upregulated proteins in EVs, we generated two scatterplots of proteins abundance: first, across 1072 proteins detected in both lung tissues and EVs (Fig. 7B) and second, across 1410 proteins identified in both type of EVs (Fig. 7C). Today, it’s well known that EVs could carry proteins from tumor cells. Thus, the goal was to pinpoint proteins upregulated in tumors vs. normal tissues that could be present in circulating EVs, which could serve as potential circulating biomarkers of lung adenocarcinoma. Among 1072 proteins detected in both tissue and EVs, 128 proteins were significantly deregulated in tumor versus non-tumor tissues. Eleven proteins were upregulated in tumor while 117 were downregulated compared to NT (Fig. 7B and 7C). In the set of 11 increased proteins in tumor, 5 appear particularly relevant (FCGBP, TUBA1C, PFKP, TNC and SRPRB), since they are detected at very high levels in EVs from TDV (more than 4 times higher than tissue levels) (Fig. 7B) and are also detected at high levels in EVs from PV (Fig. 7C).

Lung adenocarcinoma remains a leading cause of cancer-related mortality, with a high risk of relapse even after curative-intent surgery. In this context, development of liquid biopsy constitutes a critical breakthrough in cancer monitoring, providing a minimally invasive approach to identify potential tumor biomarkers and track cancer progression. Among liquid biopsy components, EVs hold significant potential as circulating biomarkers. However, as already demonstrated, EV cargo composition varies according to the sampling site (18). Indeed, a previous study has demonstrated an original miRNA signature in TDV-derived EVs, with miR-203a-3p being significantly overexpressed compared to peripheral blood EVs (18). Nevertheless, no prior study has investigated the proteomic content of EVs based on their anatomical site of origin. This study provides the first comprehensive proteomic analysis of EVs derived from TDV and PV in patients with resectable lung adenocarcinoma, combined with matched tumor tissue proteomics.

Our initial tissue proteomic profiling revealed major changes in protein expression patterns, consistent with the profound molecular remodeling that accompanies tumorigenesis. A total of 466 proteins were differentially expressed between tumor and paired non-tumor tissues, with functional enrichment analyses highlighting alterations in metabolic processes and extracellular vesicles-related pathways. Among the most upregulated proteins in tumor tissues, PES1 and TMEM165 have been implicated in tumor progression and development across different cancer types (37, 38). In addition, proteins such as PYCR1 (39), GOLM1(40–42), TUBB3 (43–45), S100β (46), CYP1B1(47), UGT1A6 (48, 49) are known to play key roles in lung cancer progression, metastasis or resistance to treatment.

These findings corroborate with known cancer hallmarks and suggest that changes in the tumor proteome may directly influence the content of secreted EVs.

EVs isolated from TDV plasma samples exhibited distinct physical and molecular characteristics compared to those from PV. They were significantly more concentrated and smaller in size. These differences are consistent with the hypothesis that sampling closer to the tumor drainage site leads to an enrichment in tumor-derived EVs, as cancer cells are known to release greater quantities of EVs than non-tumor cells (10, 11). Similarly, Navarro et al., previously reported that EVs isolated from TDV were significantly smaller than those from peripheral veins (16). Thus, TDV samples might afford a privileged source of tumor-enriched EVs in particular at early stages where systemic dilution limits the detection of tumor-derived signals in peripheral blood. In this context, proteomic analysis confirmed a clear separation between TDV- and PV-derived EVs, with 369 proteins showing differential expression. Of the 176 proteins upregulated in TDV-EVs, the most upregulated were associated with lung cancer diagnosis and prognosis (SRPRB, CCT6A, MUC5B, VDAC1, ATP1A1, Factor XII) or with other cancer types (UNC45A, SSC5D, IGHV5-51) (50, 51). This set of proteins plays a major role in tumor progression and are directly associated with poor prognosis in LUAD. Within lung cancer-related proteins, CCT6A promotes lung adenocarcinoma progression by stabilizing STAT1 and inducing metabolic reprogramming (52). MUC5B overexpression is significantly correlated with poor overall survival (OS) and progression-free survival (PFS) in lung adenocarcinoma (53, 54). EVs also carry the mitochondrial voltage-dependent anion channel 1 (VDAC) which functions reside at the crossroads between metabolic and survival pathways, therefore contributing to oncogenesis and metastasis (55). ATP1A1 is receptor Na- and K-dependent adenosine triphosphatase with ion transport activity which promotes immune escape in lung adenocarcinoma (56), The enrichment of Factor XII in tumor-derived EVs is consistent with the increased levels of polyphosphate (PolyP) in EVs from cancer patients cited in the literature. Indeed, PolyP binds FXII and is directly linked to tumor coagulopathy (57).

A key result of our analysis was the identification of SRPRB (Signal Recognition Particle Receptor Subunit Beta) as the only protein upregulated in both tumor tissue and TDV-derived EVs. SRPRB plays a central role in directing proteins to the endoplasmic reticulum and has previously been linked to cancer progression and poor prognosis in epithelial tumors (58). Its detection in TDV-EVs may reflect enhanced membrane protein synthesis and exocytosis in tumor cells (59). In a tissue-microarray study involving a broad range of normal and tumoral tissues, SRPRB was found to be upregulated in numerous epithelial cancers and was significantly associated with tumor progression and poor prognosis. Functionally, SRPRB has been implicated in the regulation of cell proliferation, apoptosis, tumorigenesis and metastasis (60). Its concurrent upregulation in both tumor tissues and TDV-EVs supports its potential role in tumor microenvironment remodeling. Altogether this result suggests that SRPRB is a relevant circulating biomarker of tumor burden.

Functional enrichment analyses revealed that proteins carried by TDV-EVs are involved in protein transport, immune system regulation, and cytoskeleton remodeling. Remarkably, over 50 proteins mapped to the Rho GTPase signaling pathway, which has been implicated in EGFR-TKI (Tyrosine kinase inhibitor) resistance and metastatic development in lung cancer (36).

Network analysis of the 369 differentially expressed proteins identified a densely connected PPI network, highlighting key hub proteins notably VDAC1 (61). All of these were upregulated in TDV-EVs and have previously been linked to poor prognosis in lung cancer.

Strikingly, we observed that a substantial subset of proteins downregulated in tumor tissue were paradoxically upregulated in TDV-EVs. We analyzed the panel of 1410 proteins commonly detected in both TDV- and PV-derived EVs. Among these, only 128 proteins were differentially expressed in tumor compared to adjacent non-tumor tissues. Interestingly, the majority of them were downregulated in tumors, suggesting that tumor cells may selectively secrete these potentially tumor-suppressive proteins via EVs as a mechanism to promote tumor progression. These included AHNAK, COL6A1, APOB, ITIH2 and MCEMP1, five proteins with previously reported tumor-suppressive roles in lung cancer (62–66). This observation suggests that tumor cells may actively extrude tumor-suppressive proteins via EV secretion, potentially as a mechanism to facilitate immune escape, survival, and metastatic dissemination. Similar processes have been described in other cancer models, such as GKN1 in gastric cancer (67), supporting the biological plausibility of this hypothesis.

Nevertheless, we find that among the upregulated proteins within the 128 proteins, 5 proteins (FCGBP, TUBA1C, PFKP, SRPRB, TNC) are both abundant in TDV-derived EVs and in the tumor. Moreover, these proteins are also common to the 2 subtypes of EVs derived from TDV or PV. They therefore could constitute excellent circulating biomarkers for detecting the presence of a tumor. In lung cancer, TUBA1C expression lung tumor-released EVs promotes astrocytes apoptosis to enable premetastatic niche formation in brain and prepare future metastasis (68). Currently, the exact role of the FCGBP protein in cancer is not clearly established, although previous work seems to suggest that this protein could be used as a valuable biomarker for the early diagnosis and prognosis of cancer (69). Conversely, recent results obtained with NSCLC cells suggest that FCGBP knock-out could promote cell proliferation, migration and invasion (70). Thus, under these conditions, it is therefore complex to integrate this protein into a biomarker signature.

Regarding PFKP, its function in lung cancer metabolism is already reported. Indeed, high expression of PFKP in lung cancer cells regulates the level of glycolysis which is directly linked to lung cancer cell proliferation (71).

Finally, TNC is significantly increased in lung cancer. It has been described as promoting lung cancer progression and a potential prognostic biomarker (72, 73).

Our pathway enrichment analysis suggests that PYCR1, TNC and SRPRB might share similar pathways related to response to stress suggesting that EVs release by the tumor could be involved in this process (74). Indeed, the function of EVs in cell stress response and resistance to cancer therapy is already known (75).

Although the evaluation of recurrence-associated biomarkers was not included in the study design due to the limited number of patients who experienced recurrence during follow-up. Nevertheless, preliminary stratification of patients based on recurrence status identified distinct EV proteomic clusters in both TDV and PV samples (Additional Fig. 4). However, these findings require validation in larger cohorts with extended follow-up to characterize relevant biomarkers associated to relapse. Nevertheless, these results underscore the potential of EV-based proteomic profiling for relapse prediction in early-stage lung cancer.

Our study highlights that EVs derived from TDV exhibit a distinct proteomic profile enriched in tumor-associated proteins, providing valuable insight into the molecular landscape of the tumor microenvironment. Among these, SRPRB emerges as a promising biomarker candidate for lung adenocarcinoma. Furthermore, our data suggest that tumor cells may actively export tumor-suppressive proteins via EVs, potentially contributing to disease progression. These findings pave the way for future studies aimed at validating EV-based biomarkers for early relapse detection and therapeutic monitoring in lung cancer.