Visualizing extracellular vesicles in cancer: from biogenesis to theranostic applications.

Introduction
Extracellular vesicles (EVs) represent a class of naturally occurring biological nanoparticles, typically ranging from 30 to 150 nm in diameter, which are encapsulated within a lipid bilayer and released into the extracellular milieu by virtually all cell lineages. Encapsulated within their lipid bilayer, these nanovesicles contain a heterogeneous repertoire of bioactive macromolecules, encompassing specific proteins, nucleic acid fragments, and metabolic constituents, which empower them to orchestrate cell-to-cell signaling by transporting bioactive materials to acceptor cells via fusion or endocytosis. EVs are typically categorized into small EVs (sEVs, historically termed exosomes), microvesicles, and apoptotic bodies, which are delineated by their divergent biogenic pathways and characteristic dimensional ranges. Ubiquitous within biological fluids, EVs are essential for both physiological and pathological processes [1, 2]. For example, tumor-derived EVs exhibit intrinsic homing capabilities that facilitate cancer metastasis and immune evasion [3–7], while EVs derived from antigen-presenting cells (APCs) exhibit lymph node-targeting properties for immune regulation [3, 4, 8, 9]. Beyond their biological significance, EVs have garnered substantial attention as endogenous nanocarriers, primarily driven by their superior safety profiles, capacity for immune evasion, and intrinsic homing capabilities [10, 11], offering distinct advantages over synthetic nanoparticles. Consequently, advancing methodologies to track EV biodistribution and function through imaging technologies has become pivotal for both understanding their roles in tumor progression and optimizing their therapeutic applications.
Although conventional interventions including surgery, radiotherapy, chemotherapy, immunotherapy, and targeted agents have progressed significantly, clinical outcomes are still hindered by persistent obstacles. Current diagnostic platforms often suffer from insufficient sensitivity which limits the early identification of tumors. Additionally, the complex heterogeneity of the neoplastic milieu, coupled with robust immune evasion strategies, frequently undermines the potency of therapeutic interventions. To overcome these concurrent limitations, EVs leverage their intrinsic ability to transport cell-specific molecular cargos. This unique capability allows them to function simultaneously as sensitive biomarkers for minimally invasive diagnosis and as engineerable vectors for targeted therapeutic delivery [12–14]. For instance, EV-based liquid biopsies can identify tumor-derived genetic alterations or membrane proteins [15–18], whereas engineered EVs loaded with chemotherapeutics or immunomodulators enable precision therapy [6]. Importantly, noninvasive imaging of EVs in vivo is vital for deciphering their spatial and temporal dynamics, monitoring drug release profiles, and validating targeting specificity.
This review offers a systematic exploration of the interconnected landscape between EV biology and imaging innovation in cancer research [19]. To illuminate the mechanisms of EV secretion as well as their multifaceted roles in tumor development, we provide a detailed examination of contemporary EV isolation approaches, labeling techniques, and sophisticated imaging platforms. We outline the biogenesis and functional diversity of EVs originating from a broad range of cellular sources and environmental cues (e.g., hypoxia, acidosis, and radiation) [20, 21]. We subsequently conducted a thorough evaluation of current EV isolation techniques and labeling approaches, including endogenous modifications such as fluorescent protein tagging and exogenous labeling techniques such as lipophilic dyes and radiotracers. Herein, we critically examine the emerging paradigm of EVs in cancer theranostics, bridging the gap between diagnostic precision and therapeutic intervention. Moreover, we conduct a comparative assessment of diverse visualization platforms such as fluorescence, bioluminescence, Magnetic resonance imaging (MRI), and Positron Emission Tomography/ Single-Photon Emission Computed Tomography (PET/SPECT) to delineate their specific strengths in terms of resolution, detection sensitivity, and translational potential. Finally, we address the unresolved challenges in EV tracking, including off-target labeling and signal quantification, and propose integrative solutions to bridge the gap between preclinical findings and clinical translation. Through a critical integration of these findings, we aspire to optimize EV-driven platforms for navigating the challenges of cancer theranostics, outlining both the current state-of-the-art and prospective avenues for next-generation applications. To bridge the gap between biological potential and clinical reality, this review proposes a ‘Visualization-Centric Framework.’ Unlike traditional reviews that treat imaging as a passive observation tool, we argue that advanced visualization acts as an active validation filter in the EV pipeline. We systematically integrate three core dimensions: (1) Biogenic Modulation, exploring how tumor microenvironment (TME) factors like hypoxia and acidosis generate distinct ‘imaging signatures’ on EVs; (2) The Integrity Check, utilizing dual-labeling and bio-orthogonal chemistry to distinguish true EV biodistribution from artifactual dye leaching; (3) Theranostic feedback, demonstrating how real-time imaging readouts (e.g., Photoacoustic or PET signals) actively guide the timing and localization of therapeutic triggers. This framework provides a roadmap for transitioning EVs from phenomenological curiosities to quantitatively validated nanomedicines.

Relationships between tumor biology and EVs
Driven by oncogenic stimuli, normal cells undergo malignant transformation through mutations in proto-oncogenes and oncogenes, accompanied by abnormal physiological alterations. Cancer cells adopt a range of mechanisms to evade immune detection, thereby promoting incremental tumor dissemination and metastasis [22]. During metastasis, circulating cancer cells seek out and colonize distant organs with a permissive immune microenvironment (“the soil for the seed”) [26]. At these secondary sites, they orchestrate the formation of a supportive TME, acquiring invasive traits that fuel the development of metastatic lesions [23]. Intercellular communication is pivotal throughout tumorigenesis, invasion, and metastasis. Crucially, primary tumor cells secrete factors, including EVs, that precondition distant organ sites (Fig. 1A). This process entails the recruitment and reprogramming of resident stromal cells to cultivate a permissive microenvironment, which is indispensable for the viability and subsequent outgrowth of metastatic seeds [24–26]. Given that metastatic dissemination remains the principal determinant of mortality in solid malignancies and acts as a key driver of clinical relapse, attaining a thorough understanding and early diagnosis of this process is critically important from a clinical perspective [27].
The biogenesis of these vesicles is initiated by the inward invagination of the plasma membrane, a process that internalizes extracellular cargos and surface proteins to constitute early endosomes [28]. Subsequently, these nascent compartments undergo a maturation process to differentiate into late endosomes, also technically referred to as multivesicular bodies (MVBs). During this phase, the limiting membrane of the late endosome buds inwardly, forming multiple specifically sorted intraluminal vesicles (ILVs) of different proteins, nucleic acids, and other cargoes, thereby leading to the formation of MVBs [29]. The fate of MVBs depends on various proteins expressed on their surfaces. The eventual coalescence of secretory MVBs with the plasma membrane results in the secretion of ILVs, which constitute the exosomal population of EVs [30]. Conversely, a distinct subpopulation of MVBs is routed toward the lysosomal compartment, culminating in their fusion and subsequent cargo catabolism [31]. Figure 1B depicts the specific biogenesis pathway. Table 1 provides a comprehensive overview of EV subtypes, their phenotypic traits, and molecular markers.

Acting as pivotal shuttles for molecular information, EVs orchestrate the entire spectrum of tumorigenesis and metastatic dissemination. Tumor-derived EVs contain information about early malignant progression and tumor metastasis, replete with pro-tumorigenic cargo that drives uncontrolled cell replication and facilitates metastatic dissemination, thereby accelerating the trajectory of malignant progression [48, 49]. EVs actively sculpt the neoplastic landscape through the horizontal transfer of a heterogeneous payload (comprising proteomic and genomic factors), thereby driving sustained angiogenesis, subverting immune surveillance, and engineering a permissive soil for malignancy. It potentiates the invasive phenotype and motility of malignant cells, ultimately culminating in the establishment of distant metastatic foci [48, 50–56]. Specifically, the enrichment of Programmed Death-Ligand 1 (PD-L1) on the vesicular surface enables EVs to subvert immune surveillance, thereby creating a permissive environment for systemic dissemination [57, 58].

Microenvironmental regulation of EVs biogenesis and function
EVs are defined as endogenous biological nanoparticles bounded by a lipid bilayer, which serve as fundamental vehicles for orchestrating intercellular signaling networks. The release of EVs and their intercellular signaling roles are intricately regulated by variations in the cellular microenvironment [59–61]. Ubiquitously secreted by diverse cell lineages, these biological nanoparticles encapsulate a heterogeneous payload of macromolecules, including functional proteins, specific lipids, and genetic materials [59, 62, 63]. Functioning as pivotal mediators of molecular cross-talk across both homeostatic and pathogenic landscapes, EVs recapitulate the biological status of their donor cells. Under laboratory conditions, to obtain EVs with specific functionalities, researchers often induce differential EV secretion by modulating the cell culture environment. Five commonly used induction techniques include hypochlorite, which alters the cellular metabolic state via oxidative stress, thereby affecting the proteomic characteristics of EVs [64]. Ionizing radiation activates DNA damage response pathways, prompting cells to secrete EVs containing repair-related miRNAs [65]. Ultraviolet radiation induces cellular stress responses, leading to upregulated expression of surface markers on EVs (e.g., HSP70) [66]. Hydrogen peroxide (H₂O₂): H₂O₂ regulates the secretion rate and cargo loading of EVs via reactive oxygen species (ROS) signaling pathways [67]. Hypoxia: Hypoxic stress acts as a potent catalyst for EV biogenesis, resulting in a substantial surge in secretion rates (exemplified by a 2– to 3-fold elevation in head and neck squamous cell carcinomas). Furthermore, this oxygen-deprived niche upregulates the sorting of angiogenic drivers like HIF-1α, thereby generating EVs with distinct functional profiles dictated by their environmental context [68–70]. For example, EVs secreted under hypoxic conditions promote tumor microenvironment remodeling, whereas EVs induced by oxidative stress may enhance immunomodulatory functions [67, 69]. This characteristic allows researchers to tailor EV production environments according to therapeutic needs, such as antitumor therapy, tissue repair, or immunomodulation [64, 66]; further details can be found in Fig. 1C. Notably, microenvironmental stimuli may also alter the membrane protein composition of EVs, thereby affecting their targeted delivery efficiency. Collectively, these observations lay a solid theoretical foundation for the rational design of next-generation targeted nanotherapeutics [71, 72].

Acidic microenvironment modulates EV secretion and cargo
During the malignant transformation of normal cells, their microenvironment undergoes significant acidification to an acidic range of pH 6.0–6.8 [73]. This acidic milieu induces increased EVs secretion by tumor cells and radically alters both cargo composition and biological function, thus redefining EV-mediated intercellular communication [74, 75]. Acidic tumor-derived EVs are highly enriched in oncogenic components, including miR-424-5p, specific proteins, and lipids, which collectively drive significant reprogramming of recipient cell functions [76]. Illustratively, melanoma-secreted vesicles have been shown to shuttle miR-424-5p into endothelial cells, where it targets and downregulates LATS2, thereby fueling an angiogenic program characterized by enhanced cell motility and growth [76–78]. Furthermore, the intercellular transfer of specific payloads, notably miR-211–5p, allows highly aggressive melanoma populations (e.g., POL) to confer metastatic traits upon their poorly metastatic counterparts (e.g., OL) through an EV-dependent mechanism. Furthermore, tumor acidosis acts as a critical co-factor that potentiates the pro-metastatic efficacy of EVs. This is achieved by triggering pH-responsive signaling cascades, notably through the engagement of proton-sensing G protein-coupled receptors such as GPR4 [73]. Collectively, these processes propel tumor cell migration and the acquisition of mesenchymal traits epithelial–mesenchymal transition (EMT), while simultaneously priming distant organs for subsequent colonization [79–81]. Excessive EV secretion establishes a vicious cycle through sustained activation of proangiogenic and immunosuppressive signaling pathways, thereby accelerating distant organ colonization in patients with melanoma and ultimately leading to dramatically increased metastatic potential [78, 82–86]. Significantly, beyond their active role in driving malignant evolution, these nanovesicles encapsulate unique multi-omics profiles (including specific proteins and nucleic acids), thereby positioning them as ideal targets for non-invasive diagnostic and prognostic stratifications [87, 88].

Oxidative stress and hydrogen peroxide-Mediated EVs Modulation
High-concentration H₂O₂ induces cell death through programmed cell death pathways, including ferroptosis and necroptosis [89, 90], whereas low-dose H₂O₂ may activate the NRF2 antioxidant pathway [91] or RpoS stress regulators to induce oxidative stress tolerance [92]. Studies have revealed that EVs released under oxidative stress play pivotal regulatory roles: (1) they increase the cellular detoxification capacity through the delivery of the mitochondrial protein Tom40 [93]or nucleic acid molecules such as miR-188-3p [94]; (2) Delivery of NRF2 via EVs serves to abrogate the phosphorylation of the NF-κB p65 subunit, thereby dampening the subsequent assembly and activation of the NLRP3 inflammasome [91], thereby mitigating inflammatory responses. At the tissue repair level, high glucose (HG)-induced endothelial EVs trigger DNA damage, yet stress-derived EVs simultaneously promote angiogenesis by modulating cell adhesion pathways [94, 95]. Notably, Downs et al. demonstrated that H₂O₂-stimulated EVs exhibit dual effects: they amplify inflammation by transporting proinflammatory factors (e.g., IL-6 and TNF-α) [91, 96], concurrently cultivating a regenerative niche by skewing macrophages toward the anti-inflammatory M2 phenotype or by deploying antioxidant payloads (e.g., superoxide dismutase) [97]. This paradoxical effect correlates with H₂O₂ concentration gradients, low concentrations (< 100 µM) predominantly activate the NRF2/KEAP1 antioxidant axis [98], whereas high concentrations (> 500 µM) exacerbate inflammatory cascades through the ROS‒NF‒κB axis [99]. Mechanistically, EVs protective functions involve multidimensional regulation: (1) direct ROS scavenging (e.g., catalase-loaded EVs nanoformulations) [100]; (2) modulation of oxidative stress/angiogenesis-related genes via circRNA/miRNA molecular sponges (e.g., the circVRK1/miR-150-5p axis) [101]; (3) altered membrane permeability to accelerate H₂O₂ degradation (e.g., AI-2-regulated bacterial tolerance mechanisms) [102]. Collectively, these insights substantiate the rational design of next-generation nanotherapeutics derived from EVs aimed at restoring redox homeostasis [103].

Hypoxia-Driven EV-Mediated angiogenesis and immunosuppression
Exponential proliferation of tumor cells drives metabolic reprogramming, notably glycolytic enhancement (Warburg effect), including augmented glucose uptake, lactate secretion, and enhanced fatty acid synthesis to meet energy and biosynthetic demands [104]. This aberrant metabolic activity drastically depletes local oxygen, rendering the TME progressively hypoxic (typically < 1% O₂ concentration) [105]. In response to hypoxia, tumor cells markedly increase EV secretion (often by 2- to 3-fold compared with normoxia), which is a key mechanism for remodeling the TME [68, 106]. These hypoxia-induced tumor-derived EVs are loaded with a distinct molecular cargo, notably enriched in Transforming Growth Factor Beta 1 (TGF-β1), hypoxia-selective miRNAs (e.g., miR-4299, miR-106a-5p), and proangiogenic factors, which mediate their functional impact [107–109].
Key mechanisms include the following: (1) Vesicles originating from oxygen-deprived malignancies drive vascular expansion by delivering TGF-β1, which functionally engages the endothelial CD105/TβR-II signaling axis [110]. (2) Control of metastatic dissemination: The vesicular transport of TGF-β1 triggers the Smad-dependent pathway, thereby orchestrating cellular phenotypic plasticity EMT and potentiating the invasive capacity of tumor cells [111, 112], whereas molecular transfer (e.g., miR-1225-5p) reprograms distant organ microenvironments to form premetastatic niches [113, 114]. (3) Hypoxia induces a 5 to 8-fold increase in TGF-β1 levels within tumor-derived EVs, thereby enhancing their immunosuppressive capacity [115], suppressing T-cell function through the following mechanisms: (a) Direct inhibition of T-cell proliferation (60–70% reduction) [116]; (b) promotion of regulatory T-cell (Treg) differentiation [107]; (c) suppression of mitochondrial oxidative phosphorylation via Smad2/3 phosphorylation [117]; (d) synergy with EV PD-L1 to inhibit CD8+ T cell activity [118]. This hypoxia-EV-TGF-β1 positive feedback loop drives tumor immune evasion and distant metastasis. Therapeutic strategies targeting this axis (e.g., the TGF-β1 inhibitor galunisertib combined with EV blockade) have the potential to increase immunotherapy efficacy [117].

Radiation-Induced EVs: bystander effects and immunomodulation
Radiotherapy serves as a central modality for malignant tumor treatment, exerting direct cytotoxicity via DNA and organelle damage through ionizing radiation [119, 120], but also profoundly modulates the abundance, composition, and intercellular communication of EVs within the TME [119, 121]. Irradiated tumor cells enhance EV secretion mechanisms, releasing EVs loaded with unique bioactive molecules (e.g., miRNAs and proteins) [122]. Post-irradiation vesicles (RI-EVs) serve as principal vectors for propagating non-targeted radiation effects (the bystander effect). By disseminating stress-associated cargo to naïve recipient cells both local and distal they instigate a cascade of genomic instability, redox dysregulation, and cellular reprogramming [121, 123, 124]. These nontargeted effects may concurrently promote tumor recurrence and normal tissue injury.
As a novel therapeutic modality, radiation-treated tumor-derived microparticles (RT-MPs) exhibit unique immunomodulatory potential. RT-MPs enhance tumor antigen presentation efficiency by upregulating MHC-I expression in nonirradiated cells [125, 126]. Experimental evidence confirms that RT-MP-treated tumor cells significantly improve CD8+T-cell recognition and cytotoxicity, a process mechanistically linked to enhanced dendritic cell-mediated cross-antigen presentation [127]. Furthermore, RT-MPs remodel the tumor immune microenvironment by modulating myeloid-derived suppressor cell (MDSC) activity [124] and augmenting T cell infiltration [127], synergizing with radiotherapy-induced immune activation. Notably, EVs function as complex regulators of radiation response with conflicting outcomes; for instance, radiation-triggered EVs carrying miR-21 facilitate protumor effects by promoting cancer cell viability and inducing immune suppression [122] while also offering therapeutic utility as nanocarriers for radiosensitizers like Cu–Au nanozymes to potentiate radiotherapy [128]. This duality suggests that combining EV inhibitors (e.g., anti-B7-H4 antibodies) with radiotherapy may represent a novel strategy to overcome radioresistance [129]. Emerging insights into the molecular dialogue orchestrated by EVs (encompassing miRNA shuttling and protein-protein interactions) substantiate the rational design of patient-specific RT-MP radioimmunotherapeutic platforms [125, 126, 130].

Nanotechnological implications: turning biological cues into engineering strategies
Understanding the microenvironmental regulation of EVs offers a blueprint for engineering “smart” nanomedicines. The acidic and hypoxic features of the TME, which naturally boost EV secretion and alter their lipid composition, can be exploited to design stimuli-responsive EV mimetics. For instance, pH-sensitive polymers can be conjugated to the EV surface to shield targeting ligands during circulation (pH 7.4) and expose them only within the acidic tumor niche (pH 6.5), thereby enhancing specificity. Furthermore, the radiation-induced “bystander effect” suggests that EVs harvested from irradiated cells (RI-Exos) possess intrinsic immunostimulatory properties, making them ideal candidates for developing radiation-guided nanovaccines. These biological insights are shifting the paradigm from using native EVs to engineering “TME-instructed” biomimetic nanocarriers.

Cellular origins and functional diversity of EVs
EVs are defined as endogenous biological nanoparticles delimited by a lipid bilayer, which serve as fundamental vehicles for orchestrating intercellular signaling networks [131, 132]. By orchestrating intricate intercellular dialogue, EVs mediate the horizontal transfer of a heterogeneous payload, comprising proteomic, lipidic, and genomic constituents (spanning both coding and non-coding species) [132–134]. The functional profile of EVs is intrinsically dictated by their parental lineage. A prime example involves vesicles originating from malignant cells, which are frequently enriched with checkpoint ligands (e.g., PD-L1) to subvert immune surveillance [135, 136], whereas, vesicles secreted by professional APCs (e.g., DCs) or mesenchymal stem cells (MSCs) are decorated with functional peptide-MHC complexes, empowering them to prime cytotoxic T lymphocytes against malignant cells [137, 138]. Such functional pleiotropy is intrinsically linked to their unique molecular architecture, which features a specific surface topology (enriched with tetraspanins like CD63 and CD81) alongside a complex luminal genetic payload, features that not only reflect parental cell status but also confer EVs with inherited functional properties [131, 133, 139]. Within the TME, EVs play dual regulatory roles: protumor effects tumor cells utilize EVs to transfer oncogenic miRNAs (e.g., miR-21) and metabolic enzymes (e.g., HK1) to normal cells, inducing EMT and metastasis [140, 141]; antitumor effects are achieved through immune cell-derived EVs (e.g., macrophage EVs), which remodel the TME via cytokine (e.g., TNF-α) delivery, thereby inhibiting angiogenesis [138, 142].
This functional duality positions EVs as a “dual functionality” in oncology, serving both as liquid biopsy biomarkers (detecting ctDNA or specific miRNAs in circulating EVs for early cancer screening [143, 144]) and as engineered targeted delivery systems (loaded with chemotherapeutics or siRNA) [134, 145]. Current research prioritizes the functional characterization of EVs from specific cellular subpopulations: Hypoxic tumor cell-secreted sEVs deliver the mitochondrial fission protein DRP1 to reprogram recipient cell metabolism and increase invasiveness [141]. Anesthetics significantly alter miRNA profiles (e.g., upregulated miR-21-5p) in the blood EVs of bladder cancer patients, suggesting implications for postoperative recurrence [146]. Figure 2A, B, and C schematically present these mechanisms. Cutting-edge engineering strategies, ranging from targeted peptide insertion on the EV surface to intracellular cargo manipulation (e.g., CRISPR/Cas9 genome editing), are transforming EVs into innovative platforms for tumor immunotherapy (e.g., chimeric antigen receptor EVs) and precision medicine [134, 145, 147].

Immune cell derived EVs: implications for antitumor therapy
The majority of EVs secreted by immune cells demonstrate substantial tumor-inhibitory potential and are thus being explored as promising nanocarriers for tumor therapy. EVs released by MSCs display both regenerative and immunomodulatory capabilities. Notably, Zhang, K. employed human placenta–derived MSC-derived EVs in Sox9-CreERT2 mice and observed that these vesicles promoted the proliferation of Sox9+ renal cells following acute kidney injury [148]. Vesicles of dendritic origin have been demonstrated to orchestrate the concentration-dependent homing of MSCs. This phenomenon relies on the upregulation of MMP-9 and the accumulation of chemotactic drivers, ultimately accelerating the regenerative process [149]. Notably, the cytotoxic potency of nanovesicles released by Natural Killer (NK) cells, pivotal effectors of the innate immune system, can be therapeutically potentiated. A case in point is the priming with Interleukin-5, which substantially augments their anti-neoplastic bioactivity [150]. By harnessing the intrinsic chemotactic homing of neutrophils to inflammatory foci, coupled with their unique capacity to traverse the blood–brain barrier (BBB), researchers have developed neutrophil-derived EVs as vehicles for chemotherapeutic drug delivery in glioma therapy [151]. Biodistribution analyses reveal a pronounced hepatic affinity for macrophage-secreted vesicles. In stark contrast, EVs originating from BMSCs or A375 melanoma cells display minimal accumulation in major clearance organs (liver, spleen, lungs), a disparity that validates the concept of intrinsic organotropism and necessitates the rational choice of donor lineages for targeted delivery [152]. Functioning as endogenous nanovaccines, vesicles released by immune effectors (such as dendritic cells) are decorated with functional peptide-MHC complexes and tumor-associated antigens, effectively priming T cell-mediated immunity. DC-EVs, for example, can greatly enhance the recognition and cytotoxicity of CD8+ T cells against tumors, thereby inhibiting tumor growth. Taken together, these data substantiate the emergence of immune cell-derived vesicles as promising bio-inspired nanoplatforms for advanced oncological interventions.

Tumor derived EVs: roles in progression and as delivery vehicles
Tumor derived EVs robustly promote tumor growth and metastatic dissemination. A seminal study by Zomer and colleagues revealed that vesicles shed by aggressive malignant clones are capable of being internalized by indolent counterparts, a phenomenon observed not only within the proximal tumor niche but also at distal anatomical locations. This process delivers specific mRNAs that promote cellular migration and metastasis [153]. Leveraging the intrinsic homing capabilities of EVs, researchers have engineered EV vectors tailored to target parental tumor cells, thereby enhancing therapeutic specificity. Autologous (parental cell-derived) EVs can facilitate preferential uptake by parent tumor cells, increasing the cytotoxicity of their cargo and enabling potential applications in personalized medicine [154]. To augment site-specific targeting, Gong and colleagues implemented a low-pH preconditioning strategy on patient-derived tumor progenitor cells. This approach generated bio-engineered vesicles that exhibited significantly improved delivery efficiency within the neoplastic niche [155]. Inhaling lung-derived EVs as drug carriers, as illustrated by Popowski, K.D., improves evasion from immune clearance, augments targeting to lung tissue, and enhances both retention and effectiveness of chemotherapeutics [156].
To harness both the high tumor antigen concentration and the homing characteristics of tumor EVs, some studies have combined them with professional APCs to construct novel hybrid EVs. For instance, Bao, P. engineered chimeric EVs derived from dendritic cell–tumor cell hybrids, thereby leveraging both the broad-spectrum antigenic repertoire of tumor cells and the robust T cell stimulation capacity of DCs [157].

Isolation and characterization of EVs
To date, a diverse arsenal of purification methodologies has been established to harvest EVs, primarily exploiting their biophysical traits (e.g., density and size via ultracentrifugation and size exclusion chromatography) or specific surface signatures (via immunoaffinity capture) [158]. Each isolation method relies on distinct principles, resulting in unique advantages, disadvantages, and variations in the size, yield, content, and function of the recovered EVs. Combining different approaches to maximize strengths and minimize individual limitations, or sequentially applying multiple methods, can substantially increase EV yield and enhance both protein and RNA content. These methodologies are illustrated in Fig. 2D, E, F, G, and H.

Ultracentrifugation-Based techniques
Ultracentrifugation is widely recognized as a premier method for isolating EVs. It enhances sample purity by employing differential centrifugation to sequentially eliminate cellular waste before harvesting EVs at high velocities. Nonetheless, the yield of EVs via ultracentrifugation typically lags behind other methods, in part due to mechanical damage caused by centrifugal forces, which can rupture EVs and cause protein leakage. The application of protective agents such as sucrose can mitigate centrifugal damage and better preserve EV integrity [161].

Extrusion-Based generation of EVs
Certain cell types secrete EVs in minimal quantities, constituting a limitation for downstream applications. To address this, researchers have developed extrusion-based approaches, such as ultrafiltration, wherein cell membranes are mechanically forced through membranes of sequentially decreasing pore size to generate nanoscale vesicles that mimic natural EVs. These engineered nanovesicles not only exhibit higher productivity but also encapsulate greater amounts of proteins and RNA compared to endogenously secreted EVs [162, 163].

Polymer-Based precipitation
Precipitation techniques use polymers such as polyethylene glycol, Prekit-Exo precipitant[164], and SubX™ [165] to change the solubility of EVs in a solvent, allowing them to precipitate, often in combination with centrifugation [161]. Compared with the ultrafast centrifugation method, the precipitation method is 6 times faster, and the EV concentration per milliliter is approximately 2.5 times greater [166].

Immunoaffinity capture
Immune capture utilizes the affinity of protein receptors on the EV membrane to bind specific proteins and uses functionalized magnetic beads [167] or a specific matrix to isolate EVs [168]. The immunoaffinity capture method can sort different subtypes of EVs according to their different membrane markers [169] and can efficiently isolate EVs in their natural state from clinical samples for CD63-positive cancer diagnostic screening [170].

Size exclusion chromatography
Alternatively, size exclusion chromatography (SEC) achieves purification by exploiting differences in hydrodynamic radius, allowing larger vesicular entities to elute preferentially while smaller contaminants are retarded within the porous stationary phase. Here, EVs traverse the column more rapidly compared to larger entities that become transiently trapped in the bead pores. For recovery and sample purity, SEC often outperforms ultracentrifugation, yielding higher particle counts and enhanced protein purity [171, 172].

Emerging technologies: microfluidics and acoustofluidics
Microfluidic technology enables high-throughput separation of EVs through sophisticated microchannel structural design and hydrodynamic control. Its core lies in the integration of sample processing, separation and purification, and detection and analysis modules onto a centimeter-scale chip and the use of the laminar flow effect, inertial focusing, or active force field (e.g., dielectrophoresis) to achieve rapid and efficient sorting of subpopulations of EVs in micrometer-scale channels [173–175]. For example, the deterministic lateral displacement (DLD) technique produces a critical separation size through a specially arranged array of microcolumns, and passive sorting of EVs can be accomplished on the basis of particle size differences alone [176]. Compared with conventional ultracentrifugation, which takes 6–8 h and is prone to vesicle aggregation, the microfluidic system reduces the separation time to less than 30 min while maintaining > 90% recovery [177, 178]. Conversely, acoustofluidic platforms represent a paradigm shift, harnessing the synergy of acoustic radiation pressure and induced streaming to execute the contact-free fractionation of EVs. When a surface acoustic wave or body acoustic wave (BAW) is applied to a microfluidic chip, a stable acoustic pressure field is formed in the microchannel, which produces a differentiated acoustic force effect by adjusting the acoustic wave frequency (usually in the range of 1–100 MHz) [179, 180]. Owing to the difference in acoustic impedance characteristics between EVs (50–1000 nm) and impurities such as cellular debris and lipoproteins, precise separation can be achieved in an unlabeled state [181, 182]. Experiments have shown that the separation purity of EVs by acoustic flow control devices can be more than three times greater than that of conventional ultraisolation methods and can maintain the integrity of vesicle membranes and biologically active molecules (e.g., miRNAs and the surface marker protein CD63) [183, 184]. In particular, standing wave acoustic tweezers can form dynamic acoustic traps through phase modulation to realize intelligent sorting of EV subtypes [180, 185]. This gentle physical separation avoids the shear force damage of traditional mechanical filtration methods (which can lead to the rupture of more than 30% of vesicles) and provides high-quality samples for subsequent functional analysis and clinical applications [186, 187]. The synergistic development of these two technologies is breaking through the bottleneck in the field of bioseparation: microfluidics provides the framework for high-throughput platforms, and acoustic wave technology empowers precise manipulation.

Labeling strategies for in vivo EV tracking
Figure 3A illustrates the application of various imaging techniques across different spatial scales, ranging from centimeters to nanometers, and their corresponding biological structures. In Fig. 3, we provide a comprehensive breakdown of visualization strategies, segregated into three distinct clusters: non-optical/scanning approaches (e.g., MRI, nuclear imaging), conventional fluorescence-based methods, and high-resolution electron-based platforms (EM and CLEM). To monitor biodistribution at the centimeter scale (10 cm to 1 cm), researchers typically deploy tomographic imaging systems, encompassing MRI as well as nuclear medicine approaches like PET and SPECT. These techniques are suitable for visualizing large-scale structures such as mouse tissue and lesion tissue. For example, MR images provide detailed anatomical information about the brain, whereas PET/SPECT images highlight metabolic activities in lesion tissues. Transitioning to the cellular and subcellular domains (1 mm down to 100 nm), optical nanoscopy has revolutionized biological imaging. This category specifically includes sub-diffraction methodologies such as Stimulated Emission Depletion (STED), Structured Illumination Microscopy (SIM), and single-molecule localization techniques like photoactivated localization microscopy (PALM). These techniques enable the visualization of cellular structures, organelles such as MVBs, and EV clusters. For example, PALM images reveal the intricate details of cellular components, whereas fluorescence microscopy (FL) images depict single EVs. At the nanoscale (10 nm to 1 nm), EM and CLEM techniques are employed to visualize ultrafine structures such as protein membrane bilayers. These techniques provide high-resolution images that reveal the detailed architecture of biological membranes. For instance, bioluminescence imaging (BLI) serves as a potent platform to interrogate molecular interplay, enabling the visualization of real-time conformational changes or assembly events. Overall, Fig. 3A demonstrates the versatility of imaging techniques in elucidating biological structures across a wide range of spatial scales, from whole organisms to individual molecules. Adopting such an integrative framework is pivotal for deciphering intricate pathophysiological mechanisms, thereby laying the groundwork for the rational design of next-generation precision nanomedicines.
Methodologies for conferring traceability to EVs can be dichotomized based on the timing of labeling: genetic encoding during biogenesis versus direct chemical functionalization after isolation. Exogenous labeling involves direct staining of the vesicle’s outer membrane, most commonly using lipid dyes. Genetic encoding involves the transfection of donor cells with reporter genes (e.g., luciferase), ensuring that the resulting bioluminescent payload is actively loaded into the vesicular lumen prior to secretion. Notably, various labeling strategies can affect the biodistribution of EVs in vivo, necessitating careful comparison and selection to ensure experimental validity. In addition to robust labeling, reliable visualization of EVs in vivo requires imaging platforms optimized for deep-tissue penetration, integrating high-resolution optics with enhanced sensitivity. This enables accurate EV tracking while minimizing photodamage, for example through the use of near-infrared spectra (700–900 nm) or multiphoton excitation systems [188]. Classical optical systems are intrinsically constrained by the Abbe diffraction limit, a physical phenomenon where the wave-like propagation of photons broadens a point emitter into a Point Spread Function, thereby restricting lateral resolution to approximately 200 nm. Superresolution microscopy (SRM) techniques circumvent this limit through distinct physical mechanisms. By exploiting the stochastic switching of photosensitive probes (such as PA-GFP or Cy-dye pairs), Single-Molecule Localization Microscopy modalities like STORM and PALM bypass the diffraction limit. These techniques separate overlapping fluorophores in the temporal domain, allowing for the precise Gaussian localization of individual emitters and the subsequent generation of pointillistic super-resolution images (~ 20 nm). However, these methods necessitate chemical fixation to suppress sample dynamics and require prolonged acquisition times (minutes to hours) for thousands of image frames, rendering them incompatible with live-cell imaging of dynamic processes.
Conversely, for the dynamic mapping of EV trajectories within physiological milieus, SIM and STED are often favored. These modalities offer superior temporal resolution, rendering them uniquely suited for capturing rapid intracellular trafficking events. SIM increases resolution to approximately 100 nm by projecting spatially structured illumination patterns that encode high-frequency information, achieving millisecond scale frame rates with low phototoxicity (excitation power ~ 10% that of STED). In contrast, STED narrows the effective excitation volume by depleting peripheral fluorescence using a donut-shaped depletion beam, producing sub-50 nm resolution. Despite its high spatial resolution, STED requires intense laser beams (> 1 MW/cm²), which can induce photobleaching and cellular damage, especially in thick tissues where light scattering necessitates even higher powers. Every SRM encounters the unavoidable challenge of the photon budget paradox.: higher excitation intensities improve localization precision but simultaneously raise photobleaching risks, thus compromising cell viability and long-term imaging. For EVs with sub-100 nm diameters, low labeling density further diminishes signal-to-noise ratio, and extended illumination in vivo causes cumulative thermal and oxidative stress. Consequently, SRM faces inherent trade-offs among spatial and temporal resolution, biocompatibility, and volumetric imaging capabilities, particularly for tracking EVs in live organisms. The schematic overview of labeling techniques is depicted in Fig. 3B.

Exogenous labeling: direct membrane staining and radiolabeling
Exogenous labeling is commonly used for in vivo tracking of vesicles, which directly mix EVs and dye. Lipid dyes, such as PKH67 [151, 189, 190], DiR [154, 191, 192], DiD [150, 151, 154, 155, 191–195], indocyanine green (ICG) [196, 197], and Cy7 [198], are commonly used as exogenous markers. However, this membrane dye tends to agglomerate, generating micelles and aggregates, which affects labeling efficiency, increases the size of EVs and affects their distribution.

Coi
nciding with advancements in clinical diagnostics, radionuclide-based imaging modalities have gained prominence for in vivo EV tracking. These techniques are distinguished by their exceptional sensitivity and depth-independent penetration, enabling the quantitative assessment of biodistribution in deep tissues. In addition to the lipophilic fluorescent dyes mentioned above, radiolabeling methods, such as 64Cu (or 68Ga) [199, 200], 99mTc-HMPAO [199, 201–203], I-131 and I-125 [204], are also used to track EVs via exogenous labeling. By engineering dual-labeled nanovesicles incorporating both radionuclides (e.g., 64Cu or 68Ga) and near-infrared fluorophores (such as Cy7), researchers achieved synergistic visualization through combined PET/optical imaging modalities. Researchers have shown that modifying the EV surface can increase dye strength and targeting. Through the hydrophobic insertion of functional lipids, Jing et al. decorated the EV surface with DSPE-PEG2000-N3 linkers and Cy7 fluorophores, thereby establishing a robust platform for targeted delivery and simultaneous imaging [200]. In addition to some traditional and commonly used radiolabels, many new types of labeling have been identified. Khan, A.A. used [89Zr] (an established isotope for the radiolabeling and trafficking of both biological and synthetic vesicles) to directly and intralaterally radiolabel EVs, achieving high radiolabeling yields. The lipophilic [89Zr]Zr (oxinate) 4 complex can pass through the EV membrane, and 89Zr can bind to metal chelating ligands within the EV, such as proteins and nucleic acids [205, 206]. 89Zirconium deferoxamine ([89Zr]Zr-DFO) is another option [198] In addition, the use of EV-coated contrast agents [207] or NIR dyes is another method for tracking EVs in vivo. EVs labeled with near-infrared dye can be tracked from the periphery to the nodes via near-infrared imaging [189]. In addition, researchers have identified novel dyes that can penetrate membranes and label EV contents, such as DNA, RNA, and protein. Chen, C. found optical superresolution to image the mir21 and mir31 of EVs via molecular beacons (i.e., MB21 and MB31), which adopt a folded secondary structure carrying a fluorophore-quencher couple, thereby suppressing fluorescence emission in the absence of target hybridization [208]. Fatty acid analogs (Bodipy FL-C16) are taken up and enter cellular lipid metabolism, eventually producing fluorescent EVs (C16-exos) [209].
Different markers have different effects on EVs, and depending on the needs of the experiment, we must judge the interference of dyes for optimal selection. In a pivotal benchmarking study, Lazaro-Ibanez and colleagues scrutinized the tracking fidelity of five distinct modalities. They juxtaposed genetically encoded reporters, comprising bioluminescent (Fluc, Nluc) and fluorescent (mCherry) proteins, against exogenous labeling strategies, specifically utilizing the lipophilic dye DiR and the radionuclide 111In. While exogenous probes like 111In and DiR offered superior sensitivity, the fusion reporter NanoLuc-CD63 was observed to compromise the physiological distribution of EVs. Thus, radioactivity stands out as a highly quantitative and reliable approach for biodistribution studies [195].

The artifact crisis: distinguishing functional cargo delivery from dye transfer
A critical bottleneck in interpreting EV biodistribution is the ‘Observer Effect,’ where the labeling strategy itself alters EV behavior or generates false positives. Lipophilic dyes (e.g., PKH, DiR, DiD), while popular, are prone to forming micelles or leaching onto serum lipoproteins (LDL/HDL) in vivo [210, 211]. This often results in a strong liver signal that reflects labeled lipoproteins rather than EV accumulation. To distinguish true EV uptake from nonspecific dye transfer, we recommend Dual-Labeling Strategies that tag both the membrane (e.g., lipophilic dye) and the cargo (e.g., fluid-phase cytosolic marker or RNA) [151, 212–214]. In this paradigm, the co-localization of signals validates vesicular integrity, while the separation of signals indicates cargo release or vesicle degradation. Furthermore, Bio-orthogonal chemistry (‘Click’ chemistry) offers a robust alternative, allowing for the covalent attachment of fluorophores or radiometals to EV surface glycan/proteins without compromising membrane stability, thereby providing a more accurate readout of biological fate.

Endogenous labeling: genetic encoding of reporters
Endogenous labeling means that the host cells have genetic modifications so that the EVs produced by them can naturally carry visual markers. GlucB, GFP, tdTomato, the Gluc-lactadherin fusion protein [148], and Rluc [215, 216] are commonly used as endogenous markers. Cells are engineered via the CRISPR-Cas9 and Cre-Loxp systems, in which marker genes are expressed on the membrane; thus, EVs can carry the marker during secretion or linked to EV extramembrane markers such as CD63 [148, 217, 218]. To visualize EV-mediated intercellular communication, Zomer exploited a fluorescence toggle switch relying on the Cre-LoxP system. Here, Cre-laden vesicles secreted by donor cells induce a permanent chromatic transition (e.g., from dsRed to GFP) in reporter cells upon internalization. Owing to color changes, they can distinguish cells that absorb or do not absorb EVs [153]. Expanding on this principle, Steenbeek utilized a fluorescence-flipping assay based on the GFP/CFP pair. This approach served as a robust tool to validate the bioactive nature of EV payloads, demonstrating successful material transfer and recombinase-mediated editing in recipient cells [219].
Depending on the microenvironmental substrate, researchers have also genetically engineered designs that promote marker gene expression only in the presence of specific substances. Lai et al. engineered a multimodal surface reporter, designated EV-GlucB, by generating a chimeric fusion between Gluc and a biotin acceptor peptide. Upon co-expression with biotin ligase, the construct undergoes specific metabolic biotinylation, yielding vesicles that exhibit intense photon emission in the presence of coelenterazine for non-invasive spatiotemporal mapping [220]. The simultaneous use of multiple fluorescent markers allows changes in the EV membrane to be observed with its inclusions. To achieve specific membrane targeting, Lai et al. employed a lipidation strategy by appending a palmitoylation motif to the amino-terminus (N-terminus) of fluorescent reporters (EGFP and tdTomato). These lipid-anchored constructs, designated PalmGFP and PalmtdTomato, enabled the precise visualization of the vesicular bilayer. This dual-function EV-RNA reporter system enables EV and EV-RNA labeling to track the RNA and communication of EVs [221].
Employing endogenous markers confers genetically encoded fluorescence to EVs, obviating the risk of dye-induced cellular damage and eliminating the issue of fluorescence leakage during live imaging.

Advanced imaging modalities for EVs tracking
Different means of labeling EVs have been discussed above, and different methods of imaging correspond to these labeling methods. The arsenal of visualization strategies available for EV tracking is diverse, primarily comprising optical modalities (FLI, BLI), MRI, and nuclear medicine techniques (notably SPECT and PET). Each methodology presents a unique set of technical specifications and trade-offs (summarized in Table 2). Among these, optical reporters and radio-isotopic tracing currently represent the gold standard for characterizing spatiotemporal kinetics. Given the high spatial resolution and deep-tissue penetration of MRI, this platform is increasingly adopted to map the spatiotemporal fate of EVs within living organisms. By quantifying biodistribution patterns and circulatory half-life, MRI facilitates a comprehensive understanding of vesicular transport efficiency and communication dynamics, thereby accelerating the development of precision drug delivery systems for cancer therapy.

Fluorescence imaging
Fluorescence imaging is a well-developed imaging modality for biological and medical diagnostic research. After EVs are fluorescently labeled, an optical microscope [151, 193]can collect the emitted light through a laser and a filter for imaging. For optical interrogation, membrane-intercalating fluorophores exemplified by the PKH family (e.g., PKH26/67), are routinely employed to stain the lipid bilayer via stable hydrophobic insertion [189, 190], DiD [150, 155, 193, 194], DiR [154, 191, 192, 195], ICG [195–197, 217]and GFP.
FLI imaging of EVs typically involves labeling the EV with a dye and then injecting it via the tail vein to observe its ability to target tumors and carry the drug or its distribution in vivo and messaging with cells. Watson, D.C., et al. used DiR-labeled EVs by tail vein injection to target their ability in tumors [191]. Zhu, L., et al. also used DID dye to label EVs secreted by NK cells to observe their enhanced tumor killing ability with or without intervention with IL-5 [150]. Zhang et al. utilized DiR-labeled nanovesicles derived from the DC2.4 cell line to map their spatiotemporal kinetics in living subjects via serial fluorescence scanning [192]. There are also many vesicle labeling methods that produce fluorescence only under specified conditions. Boussadia et al. implemented a metabolic labeling strategy by supplementing the culture media with the fatty acid analog Bodipy FL-C16. This probe undergoes biosynthetic incorporation into the host lipid machinery, resulting in the biogenesis of intrinsically fluorescent vesicles (designated as C16-exos) [209]. To construct a multifunctional theranostic nanoplatform, ICG acting simultaneously as a photoacoustic (PA) contrast agent and a sonosensitizer, was co-encapsulated with sodium bicarbonate (SBC) into EVs. This design facilitates pH-triggered, PA-guided synergistic chemo-sonodynamic intervention [197].
With ongoing technological advancements, new fluorescence labeling and imaging methodologies are urgently needed to support medical diagnostics and research. Fluorescence imaging is cost-effective and straightforward but suffers from limited tissue penetration. Conversely, fluorescence modalities operating within the near-infrared optical windows are distinguished by their superior signal-to-noise ratios. By mitigating photon scattering and endogenous autofluorescence, this approach is uniquely suited for the deep-tissue interrogation of physiological dynamics. For example, Srinivasan, S. used near-infrared dye-labeled EVs to successfully track their migration into lymph nodes via NIR imaging [189].

Bioluminescence imaging
Bioluminescence is the light produced by the enzyme-substrate reaction of luciferase via natural biological processes. Bioluminescent tracking necessitates the biosynthetic incorporation of optical reporters. Rather than direct staining, this is accomplished by transducing parent cells with lentiviral vectors carrying the luciferase gene, thereby hijacking the cellular machinery to produce self-illuminating vesicles; thus, the EVs released by these cells carry the luciferase reporter gene or are expressed on the membrane. While BLI bypasses the need for incident illumination thereby eliminating photobleaching issues inherent to fluorescence, it poses a challenge regarding signal detection. Successful imaging mandates highly sensitive detectors optimized for low-light conditions to ensure robust data acquisition. Commonly used reporter genes include Renilla luciferase Rluc [215, 216], membrane-bound Gaussia luciferase (mGLuc) [248, 248], firefly and NanoLuc luciferase [195]. To elicit optical contrast, the imaging protocol necessitates the prior injection of coelenterazine. This substrate interacts with the vesicular reporter payload to initiate the light-emitting reaction essential for tracking.
Bioluminescence imaging offers heightened specificity without interference from background autofluorescence, though it yields weaker signals and necessitates more complex cell line engineering than fluorescence imaging. Lázaro-Ibáñez et al. reported that NanoLuc-CD63 fusion altered EV systemic distribution, causing increased pulmonary aggregation, which underscores the potential physiological impacts of genetic EV modification for tracking purposes [195].

Nuclear imaging
Predicated on the detection of decaying radiotracers, nuclear imaging has evolved into a mainstream modality for diverse biomedical applications, ranging from disease identification to intervention evaluation. EVs are usually labeled with radioisotopes such as 124I, 131I, 111In-oxine [195], and 99mTc [201–203] and visualized via SPECT [195], PET [198, 205, 206], gamma cameras [201] and alternative visualization platforms to characterize the pharmacokinetic profiles of EVs in living subjects. Hong et al. demonstrated the feasibility of a radionuclide tracking platform by utilizing gamma camera scintigraphy to visualize the real-time biodistribution of intravenously administered RI-EVs. This study successfully established a robust imaging protocol for monitoring vesicles derived from thyroid cancer and NK cells, underscoring the potential of nuclear medicine for in vivo surveillance [204]. On the basis of this external vesicle nanoprobe, PET/CT and near-infrared fluorescence (NIRF) multimodal imaging were used to guide surgery in an animal model of colon cancer [200].
Currently, SPECT and optical imaging are the most commonly employed EV imaging modalities due to their low cost and accessibility. SPECT provides a broad scanning field, deep tissue penetration, and high sensitivity.

MRI
MRI is a clinical diagnostic modality that visualizes anatomical and physiological processes without ionizing radiation, providing excellent spatial and temporal resolution while mitigating radiation risks. In the realm of MRI modalities, superparamagnetic iron oxides (encompassing both SPIONs and USPIOs) serve as the dominant class of negative contrast labels for EVs [207]. MRI is distinguished by its superb spatial resolution and unlimited penetration depth without the risks of ionizing radiation. However, its broad utility is often constrained by a high detection threshold (low sensitivity) and substantial infrastructure expenses.

Multimodal imaging: integrating complementary strengths
Multimodal imaging platforms combine the strengths of fluorescence imaging, nuclear imaging modalities (PET/SPECT), and MRI to circumvent the constraints of monomodal approaches, thereby yielding multi-parametric datasets for the spatiotemporal mapping of EVs [249, 250]. Leveraging its superior detection limits and molecular discriminatory power, optical imaging serves as a powerful tool to capture the live dynamics of surface motif presentation and rapid biological processing of EVs [251, 252]. Nuclear imaging techniques such as PET/CT, which use radiotracers such as ¹⁸F-FDG, can be used to quantitatively analyze the systemic distribution of EVs and their mediated glycolytic activity changes [253, 254]. The ability to image without radiation while maintaining superior contrast allows MRI to successfully detect EVs located in deep anatomical structures and assess their dynamic transport across vessel walls [255]. The synergistic deployment of these imaging modalities facilitates comprehensive mapping of EV spatiotemporal distribution, metabolic kinetics, and functional interactions within the TME [256, 257].
For example, PET/CT combined with near-infrared fluorescence dual-modality imaging exemplifies this integration: PET exploits its high-penetration capability for whole-body screening, whereas NIRF provides subcellular resolution to track EVs from circulation to target organs [258]. Studies have demonstrated that dual-modality probes incorporating ⁶⁸Ga-labeled EV-specific agents (e.g., CD63-targeted molecules) and ICG fluorescent dyes enable PET-based quantification of the systemic biodistribution of EVs and intraoperative NIRF-guided localization of tumor margins enriched with EVs, achieving spatial resolution below 0.5 mm [259]. This “whole-body screening followed by localized precision targeting” strategy significantly enhances clinical utility, particularly in tumor therapy assessment, by simultaneously correlating metabolic activity changes (quantified via PET standardized uptake value) and anatomical remodeling (via CT/MRI image registration) [260, 261].
Furthermore, the fusion of multimodal imaging with artificial intelligence (e.g., deep learning-based image fusion) improves spatiotemporal matching precision, enabling the construction of 3D dynamic models to reveal EV-mediated intercellular communication networks [252, 256]. Imaging specificity is greatly improved through the use of stimulus-responsive multimodal probes. Since these probes produce nuclear and fluorescent signals contingent only upon interaction with biomarkers such as matrix metalloproteinases, background noise is significantly suppressed [258, 259]. For instance, nanoprobes surface-functionalized with PSMA-avid peptide ligands, ⁶⁴Cu radionuclides, and Cy5.5 fluorophores achieved dual-modality quantification of EV homing in prostate cancer models with picomolar-level sensitivity, surpassing single-modality imaging by two orders of magnitude. Collectively, these technological strides underpin the clinical translation of EV-based nanomedicines, facilitating applications ranging from early diagnostic stratification to the rigorous evaluation of therapeutic cargo delivery [257, 262].

Therapeutic applications of EVs: from diagnosis to targeted therapy
As naturally secreted nanocarriers, EVs possess intrinsic biocompatibility and can be engineered to encapsulate a diverse array of chemotherapeutic agents and acoustic-dynamic sensitizers for synergistic tumor therapy [155]. By engineering EVs to carry a payload of ICG, paclitaxel (PTX), and SBC, the research group achieved a pH-activatable platform capable of merging PA imaging with chemoacoustic kinetic treatment [197]. Zhang, D. combined metal nanoparticles with EVs that can carry chemotherapeutic drugs, which provides a platform combined with chemotherapy-photothermal therapy for cancer therapy [192]. As illustrated in Fig. 4, a two-step modular approach was adopted by Zhang: firstly, integrating biotin and FA ligands into the parental cell membrane via DSPE-PEG insertion; and secondly, achieving magnetic functionalization of the progeny EVs through streptavidin-mediated conjugation of iron oxide nanoparticles. The use of magnetic targets at cancer sites provides high concentrations of nanocarriers [263].

EVs as liquid biopsy biomarkers for cancer diagnosis
EVs transmit molecular signals linked to health and disease states and are promising as imaging biomarkers for tumor diagnosis, prognosis, and liquid biopsy [219, 265, 266]. Although tissue biopsy remains the clinical gold standard for pathological verification, its invasive nature inevitably causes mechanical disruption to the local microenvironment, posing a risk of iatrogenic tumor seeding and subsequent metastatic progression. In contrast, liquid biopsy offers a multimodal platform to decipher pathological signatures derived from biofluids. By obviating the need for invasive procedures, this modality facilitates rapid, real-time longitudinal monitoring of cancer dynamics, thereby reshaping modern clinical diagnostics [267]. In vitro modification and tracking of EVs can lead to an understanding of tumor pathogenesis, which can help us utilize EVs as effective diagnostic and therapeutic tools [196, 207]. The distribution of EVs in peripheral lymphatic vessels helps to establish a stage for a more effective immune response by the lymph nodes [189, 268]. EV aggregation in the viscera and its physiologic distribution can be verified for its efficacy and safety [152, 198, 199, 201, 203, 204, 243]. Achieving direct profiling of vesicular populations amidst the heterogeneous TME stands as a prerequisite for unravelling tumorigenic pathways and establishing novel prognostic indicators. The intrinsic metabolic signatures of EVs, specifically NAD(P)H intensity, were mapped in living subjects and patient-derived tissues, demonstrating a high predictive value for breast carcinoma. Such data highlight the efficacy of metabolic imaging in bridging the gap between basic EV characterization and diagnostic application [269]. Comprehensive proteomic profiling of serum-derived nanovesicles has unveiled a repertoire of protein signatures, validating their utility for the early detection, risk prediction, and prognostic stratification of cholangiocarcinoma patients [270]. Through a comparative proteomic interrogation of sEVs derived from diverse metabolic disease cell models, Ji identified cell-specific protein signatures. These findings demonstrate that sEVs harbor distinct molecular cargoes that serve as reliable indicators of their cellular origin. Such proteomic data can be leveraged to deconvolute the cellular sources of distinct sEV subpopulations present in the complex serum matrix [271]. Distinctions can be made on the basis of the different characteristics of circulating EV messenger RNAs (emRNAs) and tissue mRNAs. An improved emRNA detection strategy for PCa screening and diagnostic emRNA labeling was then developed on the basis of the characteristics of circulating emRNAs [272].

Image guided theranostics: from tracking to temporal optimization
The transition of EVs from passive drug carriers to active theranostic agents relies on the ability of imaging to verify biological barrier penetration and optimize therapeutic timing. Rather than merely confirming presence, advanced visualization now drives the decision-making process in treatment protocols.

Case study: visualizing BBB penetration via neutrophil-mimetics
While the BBB strictly segregates most chemotherapeutics from the brain parenchyma, Wang et al. bypassed this obstacle by developing chemotactic NEs-Exos loaded with doxorubicin. This strategy relies on the natural affinity of neutrophils for microenvironmental inflammatory cues (e.g., IL-8) to guide the active accumulation of drugs within the glioblastoma niche. The critical innovation lay in the use of Fluorescence Molecular Tomography (FMT) and IVIS spectrum imaging to provide mechanistic proof of targeting. Unlike synthetic liposomes, the NEs Exos were visualized crossing the BBB and, crucially, maintaining a high-concentration cluster within the inflammatory glioma microenvironment. This imaging data did not just track distribution; it validated the biological inheritance of LFA-1 and Mac-1 integrins from the parental neutrophils, proving that the “inflammation-drive” mechanism was functionally intact in vivo [151].

Case study: temporal optimization in sonotheranostics
A paramount challenge in nanomedicine is determining the precise window for triggering external stimuli. A multifunctional vesicular system was proposed by Nguyen Cao, featuring the simultaneous entrapment of PTX, sodium bicarbonate, and ICG. This strategy leveraged ICG as a dual-modal operative, driving sonosensitized ROS production while permitting non-invasive PA mapping. By monitoring the PA signal intensity in real-time, the researchers identified the exact time point of maximum tumor accumulation (Tmax). This allowed them to synchronize the application of ultrasound irradiation with the peak concentration of the drug payload. The ultrasound trigger subsequently catalyzed bicarbonate-driven CO₂ generation, causing endosomal rupture and cytosolic drug release. This imaging-guided temporal optimization resulted in a significantly higher therapeutic index compared to “blind” administration schedules [197].

Conclusion
Conclusively, this article synthesizes the burgeoning landscape of EV scholarship, underscoring their dichotomous nature as both pathogenic drivers of malignancy and versatile nanoplatforms for translational intervention.
We have systematically defined the complex pathways by which tumor-derived EVs orchestrate microenvironmental modulation to facilitate immune escape, angiogenesis, and metastatic dissemination. Furthermore, the discussion illustrates how external stimuli such as hypoxia, oxidative stress, and radiotherapy profoundly alter EV biogenesis and molecular composition, adding a vital layer to understanding their biological activities.
A significant portion of our discussion was devoted to the pivotal role of advanced imaging technologies. We compared the strengths and limitations of various EV labeling strategies (exogenous vs. endogenous) and imaging modalities (FLI, BLI, PET/SPECT, and MRI). This underscores the necessity of selecting appropriate techniques to accurately track EV biodistribution, homing, and functional transfer without introducing experimental artifacts. The emergence of multimodal imaging represents a paradigm shift, offering unprecedented capabilities for correlative and quantitative analysis of EV dynamics in vivo.
Ultimately, we scrutinized the translational trajectory of EVs, focusing on their dual utility as circulating biosignatures for non-invasive surveillance and as bio-engineered vectors for precision therapeutic intervention. Integration of real-time imaging is essential for validation of delivery efficiency, guiding clinical interventions, and tracking therapeutic outcomes.
Continued convergence between foundational EV biology, next-generation imaging, and innovative engineering strategies will catalyze the translation of EVs into effective, personalized cancer theranostics. Through integration of basic research, advanced imaging, and bioengineering, the field stands ready to realize the full therapeutic and diagnostic potential of EVs for individualized cancer treatment.

Discussion: Challenges and Future Perspectives
While EV nanotheranostics hold immense promise, clinical translation is hindered by critical bottlenecks in labeling, imaging, and manufacturing.

The “Observer Effect” in Labeling: Labeling often alters EV physiological identity. Lipophilic dyes can cause aggregation artifacts or increase hydrodynamic size, while bulky genetic tags may sterically hinder receptor interactions. Future efforts must pivot toward “zero-footprint” strategies, such as bio-orthogonal chemistry, to ensure stable attachment with minimal structural perturbation.

Bridging the Resolution-Sensitivity Gap: A disconnect exists between the whole-body sensitivity of nuclear imaging (PET/SPECT) and the nanometer resolution of microscopy. To resolve this, future studies should prioritize “scale-integration,” combining multiscale imaging pipelines with AI-driven fusion to link systemic pharmacokinetics with cellular uptake dynamics.

Biomolecular corona acquisition in the systemic circulation can obfuscate the functional moieties of engineered EVs, effectively altering their biological identity and promoting unintended sequestration by the mononuclear phagocyte system (MPS). Visualizing the dynamic evolution of this corona using FRET-based sensors is vital for understanding and improving in vivo targeting efficiency.

Standardization of Quantitative Metrics (MISEV2023 Compliance): To transition from descriptive to quantitative science, the field must adopt rigorous metrics. First, we recommend reporting the Particle-to-Protein Ratio (P/µg) as a standard purity index; high-quality sEV isolates typically exhibit ratios > 3 × 10¹⁰ particles/µg, enabling researchers to distinguish vesicular signals from co-isolated protein contaminants. Second, for in vivo imaging, quantification should move beyond relative fluorescence units to %ID/g (Percent Injected Dose per gram) derived from nuclear imaging (PET/SPECT) or calibrated fluorescence tomography. This standardization is essential for comparing delivery efficiencies across different EV engineering strategies.

Manufacturing and Translation: EV heterogeneity poses significant hurdles for GMP compliance and batch-to-batch consistency. Emerging microfluidic and acoustofluidic platforms offer scalable, automated solutions for high-purity production essential for clinical trials.

In conclusion, refining visualization technologies to be less invasive, quantitative, and multiscale is key to transforming EVs from biological messengers into precise, engineered tools for personalized cancer medicine. Ultimately, we propose a pipeline where imaging readouts do not just confirm delivery, but actively validate the integrity of engineered EVs and guide the timing of therapeutic interventions, transforming EVs from ‘black box’ carriers to precise, image-guided nanomedicines.