Extracellular Vesicles as Liquid Biopsy Biomarkers across the Cancer Journey: From Early Detection to Recurrence

Abstract

BACKGROUND:

Cancer is a dynamic process and thus requires highly informative and reliable biomarkers to help guide patient care. Liquid-based biopsies have emerged as a clinical tool for tracking cancer dynamics. Extracellular vesicles (EVs), lipid bilayer delimited particles secreted by cells, are a new class of liquid-based biomarkers. EVs are rich in selectively sorted biomolecule cargos, which provide a spatiotemporal fingerprint of the cell of origin, including cancer cells.

CONTENT:

This review summarizes the performance characteristics of EV-based biomarkers at different stages of cancer progression, from early malignancy to recurrence, while emphasizing their potential as diagnostic, prognostic, and screening biomarkers. We discuss the characteristics of effective biomarkers, consider challenges associated with the EV biomarker field, and report guidelines based on the biomarker discovery pipeline.

SUMMARY:

Basic science and clinical trial studies have shown the potential of EVs as precision-based biomarkers for tracking cancer status, with promising applications for diagnosing disease, predicting response to therapy, and tracking disease burden. The multi-analyte cargos of EVs enhance the performance characteristics of biomarkers. Recent technological advances in ultrasensitive detection of EVs have shown promise with high specificity and sensitivity to differentiate early-cancer cases vs healthy individuals, potentially outperforming current gold-standard imaging-based cancer diagnosis. Ultimately, clinical translation will be dictated by how these new EV biomarker-based platforms perform in larger sample cohorts. Applying ultrasensitive, scalable, and reproducible EV detection platforms with better design considerations based upon the biomarker discovery pipeline should guide the field towards clinically useful liquid biopsy biomarkers.

Introduction

Liquid biopsy is a noninvasive diagnostic test performed on body fluids to detect and monitor disease, including cancer. Liquid biopsy offers several advantages over tissue biopsies, such as noninvasive sampling, real-time monitoring, ease of longitudinal sampling, and capturing spatial and temporal tumor heterogeneity (1, 2). Blood is widely used as the liquid biopsy sample because it contains a variety of analytes suitable as biomarkers, such as circulating tumor cells (CTCs), circulating tumor DNA (ctDNA), cell-free DNA (cfDNA), tumor-educated platelets, proteins, microRNAs, metabolites, altered glycans, and extracellular vesicles (EVs) (1, 3). The current review focuses on EVs, which are a new class of liquid biopsy biomarkers with exciting potential. We summarize the role of EVs for real-time cancer detection at all stages, for screening and diagnosis, prognosis, predicting response to therapy, and identifying early recurrence. We also address the challenges associated with EV-based biomarker studies and provide guidance for navigating the discovery pipeline of EV-based liquid biopsy biomarkers.

EVs as All-Purpose Biomarkers during the Cancer Journey from Early Pathogenesis to Metastasis

CTCs, ctDNA, and EVs are currently the 3 most extensively studied liquid biopsy-based biomarkers for cancer detection and companion diagnostics (Fig. 1) (3). CTCs are malignant cells shed into circulation by primary tumors or metastases (4). cfDNA refers to all circulating DNA and encompasses ctDNA, which predominantly originates from apoptosis or necrosis of cancer cells from actively growing tumors (5). In contrast to CTCs and ctDNA, which derive specifically from cancer cells, EVs are lipid-bilayer delimited particles secreted by all cell types, and play important roles in cellular communication, immune response, and disease progression (6).

Fig. 1.

Lifecycle of liquid biopsies. Liquid biopsies based on EVs, ctDNA, and CTCs, can inform clinicians about cancer activity over time. Conceptually, fluctuations in liquid biopsy measurements may parallel tumor burden (red line). The solid line represents the hypothetical limit of detection of liquid biopsy measurement, above which the signal can be detected.

CTCs were first observed in 1869 in the blood of an individual with metastatic disease (7), while ctDNA was first noted in the serum of cancer patients in 1977 (8). The first description of EVs (although not known as “EVs” at the time) was reported in 1945 and then visualized by electron microscopy in 1967 (9, 10).

Since their early discovery, CTCs have been widely investigated as liquid biopsy biomarkers (11, 12), but their clinical utility has been limited by lack of standard isolation techniques and low circulating abundance (fewer than 10 CTCs per mL blood) (13), especially during the earliest cancer phases. A recent meta-analysis of 11 studies encompassing 1129 patients concluded CTCs may not be useful for early detection but can predict overall survival and progression-free survival (14).

Cancer cell-derived ctDNA can harbor tumor-specific genomic and epigenetic aberrations, including copy number variants and single nucleotide variants, and changes in DNA methylation patterns which reflect tumor evolution and permit dynamic disease monitoring (5). Although useful, major ctDNA limitations include low circulation levels (approximately 1500 diploid genome equivalent per mL plasma) (15), high degree of fragmentation, disproportionate representation of dying vs viable cells, and accumulation of DNA mutations from nonmalignant clonal hematopoiesis of indeterminate potential (3). Nevertheless, ctDNA has made major inroads in the clinical diagnostic space, including mutation profiling when tissue biopsies are not available, monitoring treatment response, and measuring residual disease and predicting outcomes in patients with residual disease at the end of treatment (16).

EVs are gaining interest as a new type of liquid-based biomarker to overcome CTC and ctDNA limitations, especially in early detection. EVs are abundant in circulation (approximately 10¹⁰ EV per mL plasma, albeit only a small fraction are directly released by the tumor) (17, 18) and the lipid bilayer protects multi-analyte cargos (proteins, nucleic acids, lipids, and metabolites), which serve as important analytes for biomarker discovery. Since they are continuously secreted by viable cells, EVs play an active role at each cancer stage, from early pathogenesis to metastasis (Fig. 2). EVs and other tumor-secreted factors are released into circulation by the primary tumor, priming distant sites as premetastatic niches. For instance, we have shown that gastrointestinal stromal tumors release EVs containing oncogenic protein tyrosine kinase (KIT), triggering phenotypic conversion of stromal cells into tumor-promoting cells (19) through MMP1 (matrix metalloproteinase 1) production, which promotes tumor invasion and progression.

Fig. 2.

EVs as clinical biomarkers to measure the cancer continuum. The continuous release of EVs into circulation represents the dynamic status of cancer, from early pathogenesis to metastasis.

Moreover, EVs carry cargo specific to the cell of origin, providing a real-time spatiotemporal fingerprint that can potentially be leveraged as a cancer biomarker to monitor the dynamics of disease progression. Cells secrete EVs that vary in cellular origin, density, and size (6). The 2 main pathways by which EVs are released from cells are via the formation of ectosomes/shedding microvesicles and exosomes. Microvesicles are formed by plasma membrane vesiculation or blebbing. Exosomes are released into the extracellular environment upon fusion of multi-vesicular bodies (MVBs) with the plasma membrane (20, 21). Exosomes are among the most studied and are of endocytic origin and range from 50 to 200 nm in size. The term “exosomes” is frequently used without validating endocytic origin (22). Thus, for uniformity and scientific accuracy, we exclusively use the term “extracellular vesicles” herein to represent all EV subtypes, as defined by the International Society for Extracellular Vesicles, as “[EVs are] particles naturally released from the cell that are delimited by a lipid bilayer and cannot replicate” (22).

Given the majority of these EVs will be found in the vasculature and other biofluids where they can be detected, they have the potential to serve as novel and effective biomarkers for diagnosis, monitoring, predicting therapeutic responses and assessing outcomes in cancer patients. Although the advantages of EV-based liquid biopsy make this concept promising, many technological challenges remain and will need to be explored.

Characteristics of a Good Liquid Biopsy Biomarker

Identifying biomarkers to support clinical decision-making is central to precision medicine. An ideal liquid biopsy biomarker should be specific, distinguishing the disease of interest from benign conditions, other diseases, and healthy controls (Fig. 3A and B). Broadly, disease biomarkers can be classified by application: (a) screening, (b) diagnosis, (c) prognosis, (d) predictive, and (e) monitoring (23). Screening biomarkers are used to detect early signs of disease in large asymptomatic populations. Diagnostic biomarkers help establish the disease type in symptomatic patients. Prognostic biomarkers forecast a patient’s likely outcome (e.g., progression, disease recurrence), while predictive biomarkers guide clinical treatment decisions (24). For instance, we reported that KRAS mutation status can predict disease control in metastatic colorectal cancer patients treated with EGFR (epidermal growth factor receptor) targeted therapies, such as cetuximab (25). As we continue to expand our knowledge of the biology of EVs, they are beginning to show great promise as clinically useful biomarkers.

Fig. 3.

Biomarker performance characteristics. (A), Hypothetical probability density distributions for a biomarker used to classify “healthy” (green curve) and “disease” (blue curve) populations. For a given cut-off (represented as a dotted line), the biomarker classifies any sample above the line as “disease” and any sample below the line as “healthy”; (B), Schematic description of the 4 main performance characteristics of a biomarker, i.e., sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV); (C), Sensitivity and specificity constantly compete depending on the threshold “positive” value. Receiver operating characteristic (ROC) curves are constructed by plotting the sensitivity against the false-positive rate (1-Specificity) for a given test. A perfectly random classifier ROC is a straight line (true positive rate = false-positive rate) with an area under the curve (AUC) of 0.5 (dotted line in C). As a biomarker’s discriminating power improves, the AUC moves away from 0.5 (perfectly random classifier) to 1 (perfect classifier). Thus, it is possible to rank biomarkers by AUC.

Biomarker performance can be quantitatively expressed by the area under the receiver operating characteristic curve (AUC or AUROC) (Fig. 3C). The farther away the AUC value is from 0.5, the better the classifier distinguishes between 2 populations, up to a value of 1, which signifies the perfect classifier. Besides an acceptable AUC, a biomarker test must also possess: (a) analytical validity: i.e., reliably and repeatedly produces comparable results across multiple sites and hospitals; (b) analytical sensitivity: i.e., must detect the biomarker of interest; and (c) analytical specificity: i.e., must discriminate the biomarker of interest from similar biomarkers. Ultimately, the clinical fate of a biomarker test depends on its (d) clinical validity: ability to provide true, replicable results in the clinical setting and (e) clinical utility: ability, under real-life conditions in the population of interest, to provide clinically meaningful, actionable information to the physician to tangibly improve patient outcomes.

Applications of Extracellular Vesicles as Liquid Biopsy Biomarkers

In the following sections, we discuss potential EV biomarker applications as liquid biopsies for cancer detection and management, across various stages (Fig. 4). Throughout, when available, we provide biomarker study details, including EV isolation techniques, biomarker detection methods, and performance characteristics, e.g., sensitivity, specificity, AUC, number of study patients (online Supplemental Table 1).

Fig. 4.

Overview of promising EV biomarkers for cancer diagnosis, prognosis, predicting, and monitoring. Note that EV biomarkers vary during various cancer stages, reflecting underlying changes in tumor biology. Abbreviations: MPC, metastatic prostate cancer; LPC, localized prostate cancer; PC, pancreatic cancer; BPH, benign prostatic hyperplasia; BC, breast cancer; HC, healthy control; MBC, metastatic breast cancer; NMBC, non-metastatic breast cancer; HG GBM, high-grade glioblastoma; LG GBM, low-grade glioblastoma; ES MC, early-stage multi cancer; ES HGSOC, early-stage high-grade serous ovarian cancer; MM, metastatic melanoma; ES OC, early-stage ovarian cancer; MC, multi cancer; ES&LS OC, early-stage and late-stage ovarian cancer; CA, carbohydrate antigen; IGFBP3, insulin-like growth factor binding protein; MIA, melanoma derived growth regulatory protein; sFAS, soluble fas cell surface death receptor; TIMP, tissue inhibitor of metalloproteinase; MPO, myeloperoxidase; HCG, human chorionic gonadotropin; KRAS, KRAS proto-oncogene, GTPase; FGG, fibrinogen gamma chain; MUC, mucin; APO, apolipoprotein; FOLR, folate receptor; BST, bone marrow stromal antigen; IGSF, immunogen superfamily; ITG, integrin; MYOF, myoferlin; ACSL, acyl-CoA synthetase long chain; GPC, glypican; EGFR, epidermal growth factor receptor; HER, human epidermal growth factor receptor; CD, cluster of differentiation; COL5, Collagen type V; AEBP, adipocyte enhancer binding protein; THBS, thrombospondin; SNAI, Snail family transcriptional repressor; LOX, lysyl oxidase; ACT, actin; RLN, relaxin; SLC, solute carrier; ACP, acid phosphatase; FOLH, folate hydrolase; HOX, homeobox; MSM, microseminoprotein; KLK, kallikrein-related peptidase; TMPRSS, transmembrane serine protease; STEAP, six-transmembrane epithelial antigen of prostate; CEA, carcinoembryonic antigen; PSMA, prostate specific membrane antigen; VEGF, vascular endothelial growth factor; SDC, syndecan; BCL, B-cell lymphoma; uPAR. urokinase-type plasminogen activation receptor; PD-L, programmed death ligand; IL13R, Interleukin 13 receptor; MTOR, mechanistic target of rapamycin kinase; AURK, aurora kinase; HLA, human leukocyte antigen.

SCREENING AND DIAGNOSIS

EVs can serve as potential biomarkers for detecting various cancers, including screening asymptomatic populations and confirming a cancer diagnosis. Here, we focused on studies of early-stage cancer detection, which is essential for improving patient outcomes.

Regarding screening, Zhang et al. developed a microfluidic platform of self-assembled 3D herringbone nanopatterns to detect low levels of tumor-associated EVs in plasma (i.e., 10 exosomes/μL of blood) (26). They showed that EV proteins CD24 (cluster of differentiation) (AUC = 1), EpCAM (epithelial cell adhesion molecule) (AUC = 1), and FRα (folate receptor) (AUC = 0.995) could distinguished early-stage (I/II) and late-stage (III/IV) ovarian cancers from healthy controls. The same investigators further developed the microfluidic device into a multichannel, integrated platform for ultrasensitive, multiplexed EV profiling from blood plasma. Using this device, a panel of 7 EV protein markers, EGFR, HER2 (human epidermal growth factor receptor), CA125 (cancer antigen), FRα, CD24, EpCAM, and CD9 + CD63, distinguished early-stage ovarian cancers from healthy controls with a perfect AUC of 1 and overall accuracy of 100% (95% CI, 83–100) (27). Another study tested a panel of 3 EV proteins, FGG (fibrinogen gamma chain), MUC16 (mucin), and APOA4 (apolipoprotein), to discriminate early-stage ovarian cancers from benign cystadenoma/healthy controls (AUC = 0.82; 95% CI, 0.70–0.94) with a 71.4% sensitivity and 82.9% specificity (28).

Regarding diagnosis, EVs can be leveraged for several cancers, including rare types, such as pediatric Ewing sarcoma. Samuel et al. immunoenriched tumor-derived EVs from pediatric patient plasma samples using EV membrane proteins, CD99/MIC2 and NGFR (nerve growth factor receptor), and then interrogated EV mRNA cargo for EWS-ETS transcripts, the fusion gene driving Ewing sarcoma (29). The EV count expressing EWS-ETS transcripts separated cases from controls with an AUC of 0.92 (95% CI, 0.80–1.00) and a positive predictive value of 1 (95% CI, 0.80–1.00). Additional Ewing sarcoma studies of EV protein and microRNA (miRNA) cargo are also promising. Turaga and colleagues demonstrated that EV proteins CD99, NGFR, ENO-2 (enolase), EZR (ezrin), and UGT3A2 (UDP glycosyltransferase family 3 member A2) are highly specific for Ewing sarcoma as a diagnostic biomarker using patient plasma samples (AUC = 0.79–0.97; 95% CI, 0.62–1.00); EVs were isolated by size-exclusion columns, and measured by specific ELISA assays (30). Crow et al. profiled EV miRNAs and identified 62 enriched exo-miRNAs (31). In a blinded cohort of pediatric patients with Ewing sarcoma, osteosarcoma, rhabdomyosarcoma, and non-cancer controls, the EV miRNA signature correctly detected most patients with Ewing sarcoma. Interestingly, the exo-miRNA signature failed to identify one Ewing sarcoma patient, whose tumor was found to harbor a previously unknown EWS-FLI1 fusion.

Population-based multi-cancer detection is gathering steam, especially for cancers without established screening programs (32). Since EVs contain multi-analyte cargo, they can potentially be leveraged for multi-cancer detection. Hinestrosa et al. utilized an electrokinetics platform to scalable purify EVs from plasma (n = 323) and used machine learning algorithms to identify cancer-associated biomarkers (32). A panel of 13 EV proteins could differentiate early-stage (I/II) pancreatic, ovarian, or bladder cancer from controls with 71.2% sensitivity (95% CI, 63.20–78.10) and 99.5% specificity (95% CI, 97.00–99.9, AUC = 0.95; 95% CI, 0.92–0.97). Another exciting study used a label-free EV Raman spectral signature that detects compositional differences in EV cargo due to the different vibrational and rotational modes of their chemical structures (33). The study applied artificial intelligence to samples from 520 patients with various cancers (lung, breast, colon, liver, pancreatic, and stomach) to identify a disease-specific spectral signature (90.2% sensitivity, 94.4% specificity, AUC = 0.97; 95% CI, 0.96–0.98) (34). Importantly, the signature could determine the organ of origin in approximately 3 of 4 cases. This label-free approach is rapid (1 h), requires low sample volume (10 μL), and is amenable to automation, making the assay promising for clinical applications. Overall, these data demonstrate the potential of EVs as screening and diagnostic biomarkers for early-stage cancer detection at clinically acceptable performance.

PROGNOSIS

EV biomarkers have prognostic potential to predict cancer behavior, and patient survival since EV cargo may reflect changes in tumor biology. EVs can also aid in clinical trial design by setting inclusion and exclusion criteria to risk stratify populations. Keup et al. demonstrated that EV ERCC1 and ERBB3 transcripts correlated with progressive metastatic breast cancer during longitudinal monitoring of patients (35). In a study of lung cancer patients, increase in mutant EGFR within EVs separated progressive vs stable disease (36). An 11-gene EV panel could distinguish cancer stage (localized vs metastatic) with 85% sensitivity and 75% specificity (AUC = 0.88; 95% CI, 0.78–0.98), outperforming prostate-specific antigen (PSA), which was 65% sensitive and 75% specific (AUC = 0.64; 95% CI, 0.45–0.84) (37). While monitoring treatment response, Tian et al. investigated 8 EV proteins, CA 15–3, CA 125, CEA (carcinoembryonic antigen), HER2, EGFR, PSMA (prostate specific membrane antigen), EpCAM, and VEGF (vascular endothelial growth factor), in a cohort of 112 metastatic breast cancer patients after 1 to 4 treatments (38). Using training and validation cohorts, the signature had 76.5% sensitivity (95% CI, 50.10–93.20) and 91.8% specificity (95% CI, 81.90–97.30, AUC = 0.92) for predicting partial response/stable disease vs progressive disease, outperforming the current biomarker, i.e., plasma CA 15–3, with an AUC of 0.79 (95% CI, 0.67–0.91). During longitudinal monitoring, this EV protein signature accurately captured the tumor burden in parallel with treatment response in 88.6% of cases.

Hu et al. demonstrated that EV mRNA profiles could identify breast cancer molecular subtypes as a surrogate for prognosis (39). Using an EV microfluidic affinity purification platform, the authors separated two EV sub-populations, EV^EpCAM (epithelial origin) and EV^FAPα (fibroblast activation protein, mesenchymal origin), with recovery > 80% and 99 ± 1% specificity. When tested in clinical plasma samples, EV mRNA profiling of EV^EpCAM and EV^FAPα showed 100% concordance with the tumor subtype, suggesting enriched EV mRNA cargo could serve as a prognostic biomarker. Another study of metastatic breast cancer showed tumor-derived EVs as a complementary prognostic marker to CTCs for predicting overall survival (40). Higher tumor-derived EV levels associated with worse overall survival (17.1 vs 43.3 months, P < 0.0001). In a study of triple negative breast cancer patients with residual disease post neoadjuvant systemic therapy, the expression of noncoding RNAs in plasma EVs, miR-200a-3p, miR-203a-3p, and miR-7845–5p, correlated with increased risk of recurrence following treatment (hazard ratio, HR = 1.39–2.06, P < 0.05, C-stat = 0.52–0.76, n = 79) (41). This EV miRNA profile could serve as a prognostic panel to identify high-risk patients with residual disease and aid in adjuvant treatment decisions.

In a study of glioblastoma patients (n = 105), overall survival was 19.8 months with elevated plasma EV protein IL13Rα2 (interleukin 13 receptor) vs 13.2 months with lower EV IL13Rα2 levels, which performed better than tissue IL13Rα2 expression (42). Similarly, GPC1 (glypican)-positive circulating EVs in serum were an independent prognostic marker of pancreatic cancer that could predict survival (26.2 months with low expression and 15.5 months with high expression) (43). Cumulatively, these early studies highlight the potential of EVs as prognostic biomarkers for cancer management.

PREDICTING RESPONSE TO THERAPY

EV cargos have been evaluated for their ability to predict patient response to therapy. Some of the EV biomarkers described as prognostic biomarkers in the above section can also be used as predictive biomarkers based on their ability to predict response to therapy (37, 38). In high-risk neuroblastoma patients, an EV microRNA signature of miR-29c, miR-342–3p, and let-7b predicts clinical responders and is downregulated in unresponsive patients (n = 6) but unchanged in good responders (n = 8) (44). Elevated miR-425–3p (P < 0.0001, n = 170) and miR-146a-5p (P = 0.0014, n = 100) levels were predictive of low responsiveness to cisplatin therapy and associated with cisplatin resistance in non-small cell lung cancer (45, 46). Using a comprehensive proteome analysis, Atay and colleagues showed that GIST (gastrointestinal stromal tumor)-derived EVs carry proteins (e.g., KIT, SPRY4 (sprouty RTK signaling antagonist), SURF4 (surfeit), ALIX (ALG-2-interacting protein X), PDE2A (phosphodiesterase 2A)) that could track disease burden in patients and could predict response to targeted therapy (47).

Several studies have shown the presence of PD-L1 (programmed death ligand) on EVs and their role in tumor immune evasion, which can be used to predict response to therapy and adaptive resistance (48, 49). Chen et al. have shown that metastatic melanoma releases EVs carrying PD-L1 protein, which can suppress the function of CD8T cells and facilitate tumor growth (49). Circulating EV PD-L1 protein stratified clinical responders and non-responders (AUC = 0.9184, 80% sensitivity, 89.5% specificity) with superior performance vs total plasma PD-L1 (AUC = 0.7026, 55% sensitivity, 73.7% specificity). Furthermore, the authors suggested that melanoma cell-derived EV PD-L1 could be used as a predictor of adaptive resistance in patients receiving anti-PD-L1 immunotherapy. Porcelli et al. showed that urokinase-type plasminogen activator receptor (uPAR)-positive EVs in metastatic melanoma patients can serve as biomarkers of resistance to immunotherapy with checkpoint inhibitors (50). In a cohort of 71 patients, uPAR-positive EVs released by melanoma, CD8T cells, and dendritic cells, but not total uPAR-positive EVs, were significantly higher in the plasma of non-responders vs responders (12.5% vs 0.31%, P < 0.001). These data highlight the importance of isolating tumor-specific circulating EVs to improve the performance efficiency of EV biomarkers for predicting response to therapy.

Prospective Clinical Trials That Have Evaluated the Utility of EVs as Cancer Biomarkers

EVs have been investigated in numerous trials as an exploratory end point (51–55). However, a few prospective trials evaluating EVs for cancer diagnosis or monitoring or to aid decision-making have now been completed. In a prospective two-phase adaptive clinical utility study, investigators assessed the performance of a urine EV gene expression assay, ExoDx Prostate Intelliscore (EPI test), in patients where the role of biopsy was uncertain. This included patients >50 years with prostate cancer and PSA in the gray zone (PSA 2 to 10 ng/mL) (52). A threshold value of 15.6 (incorporating expression levels of 3 genes in urinary EVs) avoided 26% of unnecessary prostate biopsies but missed only 7% of high-grade prostate cancers. EPI (AUC = 0.70; 95% CI, 0.65–0.75) outperformed standard criteria (AUC = 0.62; 95% CI, 0.57–0.67) at identifying patients with high-grade prostate cancers (P = 0.014). Using the same assay, a subsequent large community-based clinical trial at 24 sites tested EPI as a decision aid, in addition to PSA and clinical factors, to select patients for biopsies (53). A gray zone cohort as discussed above was enrolled. This study used an innovative design; all patients underwent the EPI test, but the urologists were provided the test results only in the active arm and not in the control arm. Using a previously validated score to separate high- vs low-risk groups, EPI reduced the biopsy rate by approximately 20% in the low-risk group but increased it by approximately 27% in the high-risk group, thus detecting 30% more high-grade prostate cancers in the active vs control arms. An important limitation was lack of long-term follow-up data, but the study highlights the potential for EV biomarkers in enhancing clinical decision-making.

Zhu and colleagues evaluated whether increased numbers of PD-L1-expressing EVs could predict response to neoadjuvant immunotherapy in a clinical trial of 31 patients with resectable gastroesophageal junction adenocarcinoma (54). Using nanoscale flow cytometry, PD-L1-expressing EVs were quantified and found to be independent of tumor PD-L1 expression. In patients with low tumor PD-L1 expression, pathologic complete response was 27.3% in patients with PD-L1-expressing EVs above the median (≥2.4 EV/nL) but 0% in patients below median levels (<2.4 EV/nL, P = 0.031). Although these were exploratory analyses that were not pre-specified, the results support conducting larger trials to test the predictive accuracy of EV markers for selecting patients for immunotherapy. In a prospective, single-center, phase 1/2 study, EV EGFR protein levels were measured by ELISA in patients with EGFR-mutant lung cancers receiving combination erlotinib and ruxolitinib therapy (55). EV EGFR levels were higher in cancer patients (10 to 120 pg/2 μg of EV protein) vs healthy controls (4 to 8 pg/2 μg of EV protein). In this small study of 22 patients, EV EGFR levels were reduced with prolonged duration of therapy but a definite correlation with treatment response could not be established. Much prospective work through clinical trials is needed before many of the EV biomarkers make it from bench to bedside.

Challenges in EV Biomarker Studies

We previously discussed the technical and biological issues that clinical studies evaluating EVs as biomarkers should consider (56). Here, we further discuss the challenges of EVs as a diagnostic due to mass limits and sources. Detecting circulating biomarkers shed by early lesions is limited by fundamental biological and mass transport barriers: (a) low volume of incipient tumor (diameter <5 mm), (b) biomarker variability due to tumor heterogeneity, (c) transport challenges from tumor microenvironment to circulation, (d) 5 orders of magnitude-fold dilution in blood, and (e ) short circulation half-life (57). These factors pose substantial challenges to EV biomarker studies seeking to identify and isolate circulating tumor-derived EVs among the billions of non-tumor-derived EVs.

One can question why EV levels are elevated in some disease states (58, 59), since it is biologically infeasible that these significant increases (2- to 3-fold) in circulating EVs derive from the tumor alone (60). We hypothesize that this phenomenon in cancer patients might arise from what we term “secondary release” of EVs (Fig. 5). This hypothesis posits that tumor-derived EVs exert both local and systemic effects to enhance overall EV release. Locally, tumor-derived EVs are transferred to normal stromal cells, which “educates” them into a transformed state, promoting further secondary EV release. Systemically, tumor-derived EVs diffuse into peripheral circulation via the blood stream reaching distant target cells, also “educating” them into secondary EV release. Overall, this program facilitates tumor progression, metastasis, and angiogenesis. This secondary release phenomenon can pose both diagnostic advantages and challenges. Diagnostically, it might release additional EV subpopulations in response to tumor-derived EVs, increasing the pool for biomarker discovery. However, identifying, and separating disease-associated EVs from host-derived EVs is challenging. To achieve the sensitivity required for diagnosis, especially in the early stages of cancer, when tumor-derived EVs may represent as low as 0.1% of the total EV population (61), technology is required to enrich subpopulations, focused on tumor-derived or lineage-specific EVs (62).

Fig. 5.

Secondary EV release hypothesis. Tumor-derived EVs uptake by recipient cells imparts transformation characteristic leading to a distinct population of “secondary released” EVs. Tumor derived EVs can have both local and systemic effect, as they influence cells in the vicinity of release, and diffuse in the peripheral circulation via the blood stream to reach distant target cells. The “secondary released” EVs are consequences of tumor-derived EVs, and both EV types contribute to tumor, progression, metastasis, and angiogenesis. They are distinct from normal/healthy cell-derived EVs as they represent tumor and tumor “educated” cells, and therefore, can have potential in cancer diagnosis and monitoring. (Image not to scale. EVs are enlarged compared to cells for visual clarity.)

Heterogeneous EV subpopulations pose another substantive challenge (6). Growing interest in the EV field has led to an exhaustive catalog of subtypes and classifications, which have recently been better defined but have contributed to increased heterogeneity. Nevertheless, it is difficult to track EV biogenesis to confirm specific subtypes since there are no definitively established EV markers. For instance, EV marker proteins CD9, CD63, and CD81 were thought to isolate exosomes; however, growing evidence suggests these markers are expressed across EV subtypes (22, 63). It remains unclear whether different subtypes are functionally distinct. Does it help to enrich specific subtypes (e.g., exosomes) vs entire EV populations? How do different subtypes perform as biomarkers compared to entire EV populations? Studies focused on addressing these questions can significantly advance the field. Until the field matures in terms of understanding individual EV subpopulations, one has to remember that different isolation techniques enrich different EV subpopulations (22, 64). Therefore, it is critical to use a proven isolation technique and remain consistent while evaluating EVs for a particular indication. The gold standard, differential ultracentrifugation, has major limitations and is being replaced by other techniques, which are more reproducible and scalable, as discussed further in the section on future directions.

Sample preparation can also add variability in EVs. For blood-derived EVs, blood collection procedures, sample medium (i.e., plasma, serum), and anticoagulant type can greatly influence the quantity and characteristics of EV populations (63, 65). The physical force associated with blood draws can activate platelets, releasing platelet-activated EVs. It is recommended to discard the first 1 to 3 mL of blood because of the activating effects of pressure (66). The clotting process during serum preparation similarly releases large amounts of activated platelet-derived EVs, which dilutes the EVs originating from cells or tissues of interest. Plasma is therefore preferred over serum for biomarker studies due to the low extent of platelet EV contamination (67). Plasma preparation from whole blood requires anticoagulants (ACD [anticoagulant citrate dextrose], citrate, EDTA, heparin), though it remains unclear which is best. One study reported that ACD and EDTA inhibit platelet activation better than citrate (68), while heparin is not recommended for functional EV studies as it can block EV uptake by target cells (66). In addition, EV cargo in circulation can be influenced by a person’s daily life. Exercise/physical activities have been shown to trigger rapid EV release (69). As such, biomarker studies need to carefully consider these types of factors, like in any clinical study, at the time of sample collection. Standard operating procedures for collecting and banking EVs in clinical practice are recommended to minimize variability.

EVs as biomarkers are subject to inherent difficulties related to biomarker work since the biomarker discovery pipeline is fraught with challenges (70). Despite an enormous number of studies focused on biomarkers, few are successfully translated (71, 72). One analysis estimated less than 0.1% of identified markers meet the level of clinical rigor needed to impact patient care (73). Thus far, one EV-based biomarker test–the ExoDx Prostate Intelliscore test by ExoDx^™ (74) has gained FDA Breakthrough Device Designation, though many others are currently being studied in clinical trials (online Supplemental Table 2). The low rate of clinical translation of biomarkers has resulted in several guidelines to improve study design (75), providing standards for sample collection (76), biomarker reporting (77, 78), analytical assay development, and guidelines specific to EV-based studies (22, 79, 80). Investigators interested in EV-based biomarkers should familiarize themselves with these guidelines to improve outcomes and clinical translation of the biomarker discovery pipeline (Fig. 6).

Fig. 6.

Five phases of the biomarker discovery pipeline. The Biomarker Discovery Pipeline, showing detailed stages for all 5 phases of the pipeline (81), as well as common problems leading to a slow progression of biomarkers to clinical use (70). Many guidelines and reporting recommendations have been proposed to help researchers at each stage of the pipeline. We list some of the more useful guidelines for researchers in the EV biomarker discovery field, including the Biospecimen Reporting for Improved Study Quality (BRISQ) (76), the Minimum Information About a Microarray Experiment (MIAME) guidelines (82), the 2018 Minimal Information for Studies of Extracellular Vesicles Guidelines (MISEV 2018) (22), the Transparent Reporting and Centralizing Knowledge in Extracellular Vesicle Research (EV-TRACK, an online crowd-sourcing database that shares information about EV-based studies) (83), the Standards for Reporting Diagnostic Accuracy (STARD) (78), the Clinical Laboratory Standards Institute Guidelines (CLSI), the Transparent Reporting of Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) (77), and the Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK) (79). The above outline of the Biomarker Discovery Pipeline uses the phases of biomarker research from Pepe et al. (81), with some elaboration based on the slightly different pipeline described in Zhao et al. (84), and common pitfalls and design considerations described in detail by Ioannidis et al. (70).

An ideal EV biomarker study should begin from a clearly defined phenotype, use an isolation technique consistently across experiments, validate EV biomarkers using techniques that can separate EVs from their co-isolated biomolecules, control for patient factors, ensure adequate power, and demonstrate reproducibility and scalability for clinical utility.

Future Directions: Using Evolving Ultrasensitive Techniques, EVs Could Detect Cancers Earlier Than Current Imaging Methods

Technological advancements in the EV field have led to the development of synthetic biomarkers and ultrasensitive microfluidic-based platforms for detecting single EVs. Synthetic biomarkers are emerging diagnostic tools based on bioengineered sensors, such as molecular probes or genetically encoded vectors. These sensors leverage specific properties of the tumor microenvironment, such as dysregulated enzymatic and transcriptional activity of early-stage tumors, to activate and amplify the signal from cancer cells, aiding in early cancer detection [reviewed in detail by Kwong et al. (57)].

Here, we focus on recent technological advances towards ultrasensitive EV detection. Ferguson et. al. developed an in-house fluorescent imaging-based single-EV analysis (sEVA) platform to detect mutated EV proteins in patients with stage I pancreatic cancer (n = 16) (61). Less than 0.1% of plasma EVs harbored mutated KRAS (KRAS proto-oncogene, GTPase) and p53 protein. Using modeling, they showed sEVA technology could detect pancreatic ductal adenocarcinomas with an estimated tumor size of 0.1 cm³, well below conventional clinical imaging capabilities. Other techniques have been evaluated for sEVA (85) by concentrating EVs toward the detection sensor. Kwak et al. developed an advanced EV platform via electrokinetically enhanced yield of plasmonic sensing for label-free, rapid EV detection, which could distinguish cases and controls at 100% sensitivity and specificity (n = 5 cancer, n = 3 healthy) with a sample processing time of 10 min (85). The plasmonic surface was functionalized by EpCAM and CD24 for ovarian cancer EV detection. In yet another approach, Deng et al. used a thermophoretic AND gate operation (Tango) assay to preconcentrate EVs using polyethylene glycol-enhanced thermophoretic accumulation followed by dual-aptamer recognition (86). The thermophoretic method rapidly and selectively enriched EVs in a customized microchamber through the synergistic effect of thermophoresis and convection, facilitated by a temperature gradient induced by localized laser heating (38, 87).

Other techniques that have been used for ultrasensitive EV detection include thermophoretic aptasensor (38), 3D nanostructured herringbone-based microfluidic platform (26), microfluidic resistive pulse sensing (88), 3D-structured nanographene immunomagnetic particles platform (89), affinity-selection-based nano-Coulter counter (90), pulse counter circuit-based ExoCounter (91), surface plasmon resonance (88), and nanoflow cytometry (64). A front runner has not emerged, but the above data suggest that technological improvements could permit efficient analysis of low EV concentrations in clinical samples from patients with early cancers.

Conclusion

EVs and their molecular cargo offer opportunities and promises as liquid biopsy biomarkers for real-time diagnosis of dynamic cancer status. The studies highlighted in this review, both basic discovery and applications within clinical trials, have shown the promise of EV biomarkers with robust performance characteristics to discriminate specific cancers from healthy cases, benign conditions, and other cancer types. Some of the major challenges in the field include lack of validation in well designed and large prospective biomarker studies, lack of longitudinal samples to assess the biomarker performance over time, uncertainty in the ability to diagnose disease in asymptomatic individuals, and requirement for better design considerations addressing the pitfalls associated with the biomarker discovery pipeline. Importantly, increasing the sensitivity down to the level for detecting single EVs and improving the ability to detect tumor-derived EVs among heterogenous EV populations are pivotal for efficient diagnostic and prognostic applications. Toward this direction, recent technological advancements in ultrasensitive EV detection have shown promise, which could ultimately dictate the fate of EV biomarkers for clinical translation.

Supplementary Material

Supplemental Material

Supplemental material is available at Clinical Chemistry online.

Nonstandard Abbreviations:

EV

extracellular vesicles

CTC

circulating tumor cell

ctDNA

circulating tumor DNA

EGFR

epidermal growth factor receptor

AUC

area under the curve

CD

cluster of differentiation

EpCAM

epithelial cell adhesion molecule

miRNA

microRNA

PSA

prostate-specific antigen

PD-L1

programmed death-ligand 1

EPI

ExoDx Prostate Intelliscore

Human Genes:

EGFR

epidermal growth factor receptor

ERBB3

erb-b2 receptor tyrosine kinase 3

ERCC1

ERCC excision repair 1, endonuclease non-catalytic subunit

FL1

Fli-1 proto-oncogene, ETS transcription factor

KRAS

KRAS proto-oncogene, GTPase