Introduction
Immune checkpoints are regulatory pathways that maintain self-tolerance and modulate the amplitude and duration of immune responses. In cancer, tumors exploit these mechanisms to evade immune surveillance. The most well-studied checkpoints include the Programmed Death 1 (PD-1)/ Programmed Death Ligand 1 (PD-L1) antigens and Cytotoxic T-Lymphocyte Antigen 4 (CTLA-4/CD152), which suppress T-cell activation when engaged by their ligands. Tumor cells often overexpress PD-L1 to inhibit cytotoxic T-cell activity in the tumor microenvironment (TME) [1]. Other emerging checkpoints, such as Lymphocyte Activation Gene-3 (LAG-3), T-cell Immunoglobulin and Mucin-domain containing-3 (TIM-3), and T-cell Immunoreceptor with Ig and ITIM domains (TIGIT), also contribute to T-cell exhaustion and immune escape [2]. By blocking these inhibitory pathways with immune checkpoint inhibitors (ICIs), anticancer immunity can be restored, enabling durable tumor control in responsive patients.
The advent of ICIs has revolutionized the treatment of many cancer types. However, only a subset of patients responds to these therapies. To optimize patient selection, there is an urgent need for reliable predictive biomarkers. While tissue-based expression of PD-L1 is currently the only established biomarker in clinical routine diagnostics, this marker has limitations including tumor heterogeneity, expression plasticity and limited tissue availability. Additionally, PD-L1 may be shed from tumor cells and redistributed to extracellular vesicles (EVs) or other cell populations, potentially confounding immunohistochemical detection on both tumor and immune cells. Liquid biopsy-based markers, including circulating tumor DNA (ctDNA) and EVs, could therefore help address these limitations.
EVs represent a heterogeneous group of membrane-enclosed particles, and a universally accepted classification system has yet to be established. Traditionally, EVs secreted by living cells have been categorized based on their size using operational terms such as small EVs (sEVs, ≤150 nm) or large EVs (lEVs, 100–1000 nm) which correspond to pellets collected at different g-forces during ultracentrifugation, the most frequently used method for EV isolation. From a biological point of view, two types of EVs can be distinguished based on their biogenesis: (1) ectosomes, which originate from the outward budding of the plasma membrane, and (2) exosomes, which arise from intraluminal vesicles formed via inward budding of the endosomal membrane and are secreted once a multivesicular endosome fuses with the plasma membrane. Unless otherwise stated, this article will primarily use the operational classification since the exact biogenesis mechanism has not been investigated in most studies.
EVs contain proteins, nucleic acids, and lipids that reflect the phenotype and functional state of their cells of origin. EVs play central roles in tumor biology, particularly in immune modulation, angiogenesis, and formation of pre-metastatic niches [3]. Using standardized methods such as ultracentrifugation, size-exclusion chromatography, or immunoaffinity capture, EVs have been successfully isolated from all human body fluids. To date, most studies on EVs as cancer biomarkers have focused on blood, hypothesizing that cancer-derived EVs are detectable once the tumor has gained access to the circulation. Modern omics technologies enable detailed profiling of EV cargo and have resulted in the identification of several EV-associated cancer biomarkers, including PD-L1 [4, 5].
Biological Background and Mechanisms
PD-L1 is a key immune checkpoint ligand leveraged by cancer cells to evade immune detection by binding to PD-1 on T-cells. Beyond surface expression on tumor cells, PD-L1 is actively packaged into EVs and released into the circulation. These EV-associated PD-L1 molecules can travel systemically, binding PD-1 on T-cells at distant sites, including lymph nodes, thereby exerting immunosuppressive effects beyond the local TME [4, 6]. Exposure to interferon-γ (IFN-γ) further enhances PD-L1 incorporation into exosomes, reinforcing the immunosuppressive milieu by suppressing CD8+ T-cell function [4, 7–9]. IFN-γ/ Signal Transducer and Activator of Transcription 1 (STAT1)-dependent PD-L1 export onto exosomes seems to require a proper trimming and maturation of glycosylated PD-L1 in the Endoplasmic Reticulum (ER) for its export to the cell surface and subsequent entry into the endosomal pathways by endocytosis [10]. Supporting the endosomal origin of PD-L1 on sEVs, epidermal growth factor receptor (EGFR) mutations and subsequent tyrosine kinase inhibitor (TKI) treatment have been shown to elevate the release of PD-L1 on sEVs dependent on Hepatocyte Growth Factor-Regulated Tyrosine Kinase Substrate (HRS)- and ALG-2-Interacting Protein X (ALIX), two critical factors for exosome release [11]. Regarding the differences in PD-L1 expression between EV subtypes, the expression of PD-L1 on lEVs has been neglected so far [6, 12] or was described as significantly lower compared to sEVs in the case of melanoma [13]. Studies from our own group have revealed that PD-L1 is also readily detectable on lEVs (>200 nm diameter), presumably being ectosomes, at levels that can highly surpass exosomal expression, at least in non-small cell lung cancer (NSCLC) [14], suggesting heterogeneous mechanisms governing export of PD-L1 molecules onto EVs. Additionally, current evidence reveals substantial differences not only between tumor types but also across individual studies, likely attributable to the lack of standardized definitions and isolation protocols for EVs.
EV-mediated immunosuppression is not limited to direct PD-L1 cargo. For instance, colorectal cancer (CRC)-derived sEVs can induce PD-L1 expression on tumor-associated macrophages, thereby contributing to a suppressive TME [15]. In gastric cancer, sEV-associated PD-L1 has been implicated in expanding myeloid-derived suppressor cells (MDSCs) via the interleukin 6 (IL-6)/STAT pathway, leading to broad immune evasion and poorer patient outcomes [16]. Similarly, research on glioma subtypes has shown that tumor cells lacking Phosphatase and Tensin Homolog (PTEN) secrete sEVs rich in PD-L1, linking oncogenic Phosphoinositide 3-Kinase (PI3K)/ AKT serine/threonine kinase (AKT) signaling to immune suppression [17]. In head and neck squamous cell carcinoma (HNSCC), PD-L1-bearing tumor-derived sEVs have been shown to drive the differentiation of regulatory T-cells and pro-tumoral M2 macrophages, creating a positive feedback loop that sustains immune suppression. Notably, removing PD-L1 from these vesicles reduced tumor growth and improved immune control in preclinical models [18]. Collectively, these data illustrate that EV-associated PD-L1 is not just a passive marker but an active participant in shaping immune escape and resistance to immunotherapy [19].
Clinical application as a prognostic and predictive biomarker
The clinical relevance of EV-associated PD-L1 is underscored by its association with tumor characteristics such as stage, size, nodal involvement, and metastasis [20]. Multiple studies have validated the prognostic utility of circulating PD-L1+ sEVs in diverse cancers including diffuse large B-cell lymphoma [21], CRC [22], gastric [23], pancreatic [24], and lung cancer [25, 26].
Beyond prognostic value, PD-L1+ EVs also hold predictive potential in immunotherapy [27]. In a prospective study of melanoma patients, PD-L1 on sEVs proved to be more sensitive and dynamic than tissue-based PD-L1. It was consistently present in all patients, more abundant than soluble PD-L1, and exhibited immunosuppressive activity comparable to tumor cells. Importantly, changes in sEV PD-L1 levels during ICI treatment correlated with therapeutic response and clinical outcome, allowing for risk stratification [28]. Further supporting this, Chen et al. demonstrated that melanoma cells secrete PD-L1+ exosomes in response to IFN-γ, which suppress CD8+T-cell function and reflect tumor burden. Clinically, baseline exosomal PD-L1 levels were elevated in non-responders to anti-PD-1 therapy, while early increases (within 3-6 weeks) during treatment predicted response with high accuracy [4]. Additionally, a recent study investigated the role of PD-L1+ and PD-1+ sEVs in metastatic melanoma patients undergoing ICI therapy. In a discovery cohort (n=71) and a validation cohort (n=22), higher levels of PD-L1+ sEVs (from melanoma and CD8+ T-cells) and PD-1+ sEVs (from all sources) were associated with non-response to therapy and worse survival. Multivariate analysis confirmed that tumor-derived PD-L1+ sEVs and T- and B-cell-derived PD-1+ sEVs were independent biomarkers of resistance [29]. Collectively, these findings highlight circulating EV-associated PD-L1 as a promising dynamic and non-invasive biomarker for monitoring and predicting immunotherapy efficacy, particularly in melanoma.
A study in patients with advanced NSCLC evaluated sEV-associated PD-L1 as a predictive biomarker for ICI therapy. In both retrospective (n=33) and prospective (n=39) cohorts, increases in sEV PD-L1 levels during treatment (at 9 weeks) were associated with non-response, as well as shorter progression-free and overall survival. In contrast, tissue PD-L1 expression failed to reliably predict treatment outcomes [30]. Beyond protein content, another study evaluated sEV PD-L1 mRNA levels as a predictive biomarker for response to anti-PD-1 therapy in patients with melanoma (n=18) and NSCLC (n=8). Using digital droplet PCR, PD-L1 mRNA was measured in plasma-derived sEVs at baseline and after 2 months of treatment. Responders showed a significant decrease in vesicular PD-L1 mRNA levels, whereas non-responders exhibited stable or increasing levels, particularly in progressive disease [31]. These findings indicate that dynamic changes in vesicular PD-L1 protein as well as mRNA correlate with treatment response and may serve as a non-invasive biomarker for monitoring ICI efficacy. Results from our own group have shown that measuring plasma PD-L1+ lEVs offers predictive power for identifying ICI responders, with an overall AUC of 0.79 and an AUC of 0.91 specifically among the subgroup of patients with PD-L1-negative tumors in tissue biopsies. Here, the predictive power of the PD-L1+ lEVs was already accessible at baseline and did not require longitudinal sampling [14].
Most studies did not specify the cellular origin of PD-L1, which limits its interpretability as a tumor-specific biomarker. PD-L1 can also be expressed on non-tumor cells, such as platelets, reducing its specificity for tumor-derived EVs [32, 33]. Theodoraki et al. investigated the predictive value of PD-L1+ sEVs derived from either tumor or T-cells in HNSCC patients (n = 18). They found that high baseline levels of CD3-PD-L1+ sEVs were associated with better response to immunotherapy. In contrast, an increase in CD3-PD-L1 sEVs during treatment was observed in patients with recurrent disease, whereas a therapy-induced decrease was only seen in those who remained disease-free [34]. In our own work, we found that PD-L1 on lEVs was predominantly associated with CD45-/CD62P+ EVs, suggesting a platelet-derived origin rather than tumor cell specificity [14]. These findings underscore the importance of characterizing the cellular origin of EV-associated PD-L1 to improve its utility as a reliable and tumor-specific biomarker.
Correlation with tissue PD-L1 expression
While plasma PD-L1+ EV levels can mirror tumor PD-L1 expression [35], cancer cells often shift significant amounts of PD-L1 onto EVs [6]. This redistribution may explain inconsistent correlations observed between tissue- and EV-derived PD-L1 across studies [14, 20, 28]. Tumor heterogeneity and the presence of molecularly distinct metastatic sites also limit the predictive power of tissue PD-L1 [36]. Supporting this, circulating tumor cells often show higher PD-L1 expression than matched tissue samples [37, 38]. Despite these complexities, immunohistochemical staining of tissue PD-L1 currently remains the only FDA-approved predictive biomarker, although meta-analyses confirmed its limited ability to reliably identify responders to immunotherapy.
Therapeutic potential
Multiple preclinical studies have investigated methods to interfere with the production and release of PD-L1-containing EVs as a strategy to mitigate tumor-mediated immune evasion [19, 39, 40]. One illustrative approach involves Late Endosomal / Lysosomal Adaptor, MAPK and mTOR Activator 1 (LAMTOR1), a key regulator of lysosomal trafficking, which was found to suppress the secretion of exosomal PD-L1 in NSCLC. LAMTOR1 achieves this by binding to HRS, a component of the Endosomal Sorting Complex Required for Transport (ESCRT) machinery responsible for sorting ubiquitinated proteins into exosomes, thereby directing PD-L1 toward lysosomal degradation and reducing its incorporation into EVs. In murine lung cancer models, treatment with a LAMTOR1-derived peptide in combination with anti-PD-1 therapy led to significantly improved survival outcomes compared to either agent alone [41–43]. In a complementary strategy, Wang et al. identified Milk Fat Globule-EGF Factor 8 (MFGE8) as a factor that facilitates PD-L1 loading into EVs. Monoclonal antibody-mediated inhibition of MFGE8, when combined with anti-PD-1 treatment, successfully restored responsiveness to immune checkpoint blockade in animal models, highlighting its translational promise [44].
Pharmacologic disruption of EV biogenesis has also shown potential. Shin et al. demonstrated that sulfisoxazole, a sulfonamide antibiotic, inhibits PD-L1 secretion on sEVs and enhances the efficacy of immune checkpoint inhibitors [45]. Mechanistically, sulfisoxazole binds to the endothelin receptor A, disrupting the endosomal trafficking essential for exosome formation and promoting the lysosomal degradation of PD-L1 cargo [46]. Similarly, EP16, a moclobemide-derived compound, has emerged as an inhibitor of PD-L1 on CD63+ EVs. In gastric cancer models, combined treatment with EP16 and anti-PD-1 therapy resulted in a tumor growth inhibition rate of 69 %, exceeding the effects of either monotherapy [47]. In how far these approaches can also block PD-L1 export onto lEVs, and whether tumor cells can dynamically adapt PD-L1 export routes under treatment are questions that remain to be answered.
Advanced nanotechnology-based approaches have also been developed. For example, a tumor-targeted nanomodulator incorporating amlodipine, a calcium channel blocker, and GW4869, an inhibitor of neutral sphingomyelinase (nSMase), have been designed to suppress exosomal PD-L1 release [48]. Amlodipine reduces intracellular calcium levels, preventing calpain-mediated Beclin1 degradation and promoting autophagy of PD-L1–rich recycling endosomes [49]. Meanwhile, GW4869 blocks ceramide synthesis, thereby inhibiting exosome formation [50], while boosting ecotome shedding [51]. Still, in hepatocellular carcinoma models, this dual strategy successfully reprogrammed the immunosuppressive tumor microenvironment and reduced metastasis.
Beyond immune suppression, EV-associated PD-L1 has been implicated in chemoresistance. In CRC models, PD-L1 on sEVs was shown to promote resistance to oxaliplatin by modulating DNA damage response pathways [52]. Its secretion was driven by type Iγ phosphatidylinositol phosphate kinase (PIPKIγ) via Nuclear Factor kappa-light-chain-enhancer of activated B-cells (NF-κB) signaling. Pharmacological inhibition of PIPKIγ or NF-κB led to decreased vesicular PD-L1 release, suggesting a potential strategy to overcome chemoresistance. Additionally, widely used drugs such as ibuprofen have demonstrated the ability to reduce PD-L1 secretion in sEVs, offering practical opportunities for drug repurposing [13]. The reduction of circulating EVs carrying PD-L1 could have another benefit, as they were found to bind nivolumab, thus reducing available antibody levels for trafficking to T-cells, and identifying another mechanism of EV-mediated therapy resistance [29].
Targeted delivery strategies have also been employed to interfere with immune checkpoint signaling. Li et al. developed sEVs loaded with small interfering ribonucleic acids (siRNAs) targeting PD-L1 and CTLA-4. In CRC xenograft models, treatment with these siRNA-loaded sEVs reactivated antitumor immunity and significantly suppressed tumor growth [53, 54]. Furthermore, activated T-cells naturally released PD-1–bearing sEVs that bound to PD-L1 on tumor cells, promoting its internalization via clathrin-mediated endocytosis and reducing its availability to inhibit T-cell activation. This PD-1+ sEV-mediated mechanism has been shown to bolster antitumor immune responses and decrease tumor burden in preclinical breast cancer models.
Advantages, Limitations, Perspectives
Compared to traditional tissue-based PD-L1 testing, the evaluation of PD-L1 on EVs offers several compelling advantages that address key limitations of current approaches. Tissue biopsies typically provide only a static, localized snapshot of PD-L1 expression, which may fail to capture the spatial and temporal heterogeneity of tumors, particularly in advanced disease where multiple metastatic sites can exhibit divergent profiles. In contrast, EV-associated PD-L1 analysis can integrate signals from both primary and metastatic lesions, as well as tumor and stromal compartments, offering a more comprehensive representation of the tumor’s molecular landscape. Moreover, EV-based assays are minimally invasive, relying on blood, or other biofluids, for sampling. This enables repeated, longitudinal monitoring over the course of treatment, allowing clinicians to track real-time changes in PD-L1 expression and the evolving immunological milieu. Such dynamic information has the potential to improve patient stratification, optimize timing and selection of immune checkpoint inhibitors, and detect early signs of resistance or relapse.
However, despite these advantages, several important challenges and limitations must be addressed before EV-PD-L1 testing can be broadly implemented in clinical practice. A major barrier is the lack of standardized, validated assays for quantifying EV-associated PD-L1, with current methodologies varying widely in sensitivity, specificity, and reproducibility. Additionally, biological variability introduced by differing EV isolation and enrichment techniques can significantly affect results, complicating cross-study comparisons and clinical interpretation. To address these challenges and promote consistency across the field, the International Society for Extracellular Vesicles (ISEV) has published the MISEV guidelines aimed at standardizing methodologies and best practices for EV research [55]. There is also a pressing need for large-scale, prospective clinical trials to rigorously validate the predictive and prognostic value of EV-PD-L1 measurements, to determine optimal thresholds for clinical decision-making, and to assess their integration with other liquid biopsy biomarkers such as ctDNA or soluble tumor-associated proteins. Addressing these issues will be critical to establishing EV-PD-L1 as a robust and reliable biomarker.
Taken together, these considerations suggest that EV-associated PD-L1 holds significant promise as a predictive, dynamic biomarker capable of guiding immunotherapy strategies. With further technological standardization, analytical validation, and clinical evidence, EV-PD-L1 testing has strong potential to become an integral component of multimodal liquid biopsy approaches, ultimately advancing personalized cancer care.
Acknowledgements
MK was supported by the Deutsche Forschungsgemeinschaft (DFG project 493624047 (Clinician Scientist CareerS Münster)) and Gilead Sciences GmbH. KM received funding from the Wilhelm Sander-Stiftung (project 2022.139.1), Gilead Sciences GmbH and the Interdisciplinary Center for Clinical Research (IZKF) Münster (project Hai4/007/25).