As MSC-EV-based therapeutics continue to progress toward translational and clinical applications, upstream cell culture systems are being subjected to increasingly stringent requirements with respect to scalability, process robustness, and controllability. At the initial stage of EV manufacturing, the deliberate selection and qualification of EV-producing cell sources represent critical determinants of batch-to-batch consistency, biosafety, and functional performance of EV products [Figure 2A]. This process requires careful consideration of MSC phenotype, donor- and passage-dependent biological variability, as well as potential contamination introduced during cell isolation, expansion, or handling. Insufficient control at this upstream level can propagate variability throughout downstream processes, ultimately manifesting as heterogeneity in EV yield, molecular composition, and biological activity.

Once an appropriate EV-producing source has been established, cell expansion becomes the key enabling step for standardized and scalable EV production [Figure 2B]. Cell culture of MSCs in vitro is commonly performed using either adherent or suspension-based cell culture strategies. A traditional method of culturing MSCs on a two-dimensional (2D) planar surface such as plastic culture flasks or plastic petri dishes is still prevalent among researchers. However, it is limited in terms of surface area to volume ratio which restricts cell density and makes the scalability of MSCs ineffective; in addition, it is very labor intensive to grow MSCs in large quantities because of the higher potential for contamination associated with the continued manual handling of the MSC cultures^[61]. Culturing MSCs in monolayer format provides minimal cell-to-cell interactions; as a result, over time, MSCs under 2D monolayers have a tendency to lose their unique characteristics found in their natural living environment. Three-dimensional (3D) cultures have the advantage of replicating the in vivo microenvironmental architecture and cellular patterns in greater detail and therefore provide more physiologic relevant behaviors for the cells grown. Manufacturing advantages of 3D systems include: enhanced nutrient delivery within the system; decreased number of manual manipulation steps; and increased feasibility for manufacturing MSCs on a large scale^[62,63].

The majority of MSC culture methods using 3D cultures can be separated into scaffold-based and scaffold-free systems^[64]. Several scaffold-free methods of creating 3D cultures (such as mechanical agitation, hanging drop assemblies, low adhesion substrates, and magnetic levitation) facilitate the spontaneous aggregation of suspended cells into multicellular spheroids^[65]. These techniques help to maintain MSC phenotypes and allow for intercellular communication. However, as spheroid size increases, diffusion of oxygen, nutrients, and metabolic waste becomes progressively limited. Consequently, large spheroids are prone to developing hypoxic or nutrient-deficient cores, leading to reduced cell viability, central necrosis, and structural instability. Therefore, many scaffold-based 3D culture systems are utilized to assist in overcoming these limitations because they provide exogenous (outside) support for the adhesion and growth of the cells. Examples of common scaffolding materials used in 3D tissue engineering include Hydrogels, Fibrous matrices, Porous frames, and Microcarriers^[65]. Microcarriers, typically classified as either non-porous^[66] or porous^[67], have been widely adopted in large scale manufacturing of cells^[68]. Cells grown on solid microcarriers typically form surface monolayers, whereas cells grown on porous microcarriers have more surface area to attach to and can migrate, proliferate, and grow within the internal pore networks. Microcarrier configurations provide mechanical protection to the cells during dynamic culture conditions. Also, by using microcarrier suspension systems, large-scale cell expansion can be easily integrated with perfusion methods using closed, automated bioreactor systems, resulting in high-density cell expansion with reduced risk for contamination and improved utilization of both space and processing.

Efficient isolation and enrichment of EVs critically depend on the rational exploitation of their intrinsic physicochemical and biological attributes, including size, density, surface charge, lipid membrane properties, and surface antigens. Different isolation strategies are based on distinct principles and are therefore suited to specific applications [Table 1]. When classified according to the dominant separation mechanism, ultrafiltration (UF), tangential flow filtration (TFF), ultracentrifugation (UC), size-exclusion chromatography (SEC), Microfluidic chip (MC) and asymmetric flow field-flow fractionation (AF4) primarily exploit size differences; polymer-based precipitation (PP) depends on lipid membrane properties; affinity-based precipitation (AP) relies on surface antigens; density gradient centrifugation (DGC) separates EVs according to buoyant density; and anion-exchange chromatography (AEC) discriminates particles based on surface charge [Figure 2C].

UC has traditionally been viewed as the “Gold Standard” method for isolating EVs. The advantage of UC over other methods is its ability to separate biological nanoparticles based on their sedimentation coefficient, allowing for the enrichment of biological nanoparticles of known sizes and/or weights. Because of these differences, even EVs of similar size/mass may exhibit different rates of sedimentation when subjected to the forces exerted by the rotor. Therefore, a disadvantage of UC is that it tends to remove significant amounts of functionally relevant EV subpopulations from collections of EVs; furthermore, under certain circumstances where viscous biological fluids are present (e.g., whole blood), UC will also co-purify non-vesicular nanoparticles and protein aggregates together with EVs, thus lowering the purity of EVs isolated by UC^[69]. The application of DGC as a means for improving upon the limitations of UC is that by using either continuous or stepwise density gradients and by continually applying centrifugation force, one can have complete freedom for vesicles to migrate to their equilibrium positions based solely on the force of the centrifuge rotor. By improving the separation resolution (i.e., separating out the particles based on their density), DGC has proven capable of isolating EV subpopulations from other sources, such as those produced by mitochondria^[70]; as well as isolating both large EVs (lEVs) and sEVs^[71] and isolating non-EV nanoparticles^[69]. However, both DGC and UC are performed on devices equipped with an ultracentrifuge, have low throughput, long processing times, limited scalability, and in some instances, result in the aggregation and/or rupture of membranes due to prolonged exposure to centrifugal forces; therefore, there are concerns about the potential for DGC and UC to produce therapeutic-grade EVs^[72].

Filtration-based separation strategies offer an alternative route for EV isolation by exploiting size-dependent exclusion effects and have been extensively applied across diverse bioseparation fields^[73]. UF can be performed in either dead-end or tangential flow configurations. While dead-end filtration is susceptible to rapid membrane fouling, TFF introduces a cross-flow component that continuously removes retained particles from the membrane surface, thereby reducing clogging and maintaining stable filtration performance^[74]. Current TFF systems are typically implemented using flat-sheet membrane cassettes or hollow-fiber modules^[75]. Among these, hollow-fiber membranes provide relatively gentle and uniform flow conditions due to their linear flow paths, which helps preserve EV structural integrity during processing. Owing to these characteristics, TFF systems are capable of handling sample volumes ranging from several to tens of liters in a single operation, while achieving high EV recovery, efficient removal of soluble proteins and free nucleic acids, and substantially reduced processing times. These features make TFF particularly attractive for the isolation and concentration of clinical-grade biological nanoparticles^[76]. Accordingly, hollow-fiber-based TFF platforms have increasingly been adopted as core technologies for large-scale EV manufacturing.

MC technologies have begun to play an increasing role as a second option for separating and enriching EVs, especially in smaller format as well as diagnostic settings. Typically, these MC systems make use of microchips that combine filtration systems with specified pore sizes or multi-stage architectures. This allows for the sequential removal of both cellular components and debris and subsequently allows for the selective enrichment of EV subpopulations according to size^[77]. To assist in increasing the efficiency of the separation process and to reduce the rate of membrane fouling, dynamic flow control methods such as recirculating flow^[78], cross-flow filtration^[79] and vibrating membranes^[80] have been developed. All of these methods seek to minimize the issue of clogging of the membranes while maximizing the maintenance of mild processing conditions. MC-based EV isolation technologies combine precise fluid handling capabilities with size-selective filtering and provide a highly integrated system with minimal sample consumption and excellent reproducibility. They are especially useful when isolating EVs from small sample volumes of biological fluids.

SEC separates EVs through the exclusion effect of porous gels, with performance influenced by column height, gel type, pore size, and fraction collection range^[81]. SEC operates under relatively mild conditions and thus maintains the biophysical properties of EVs; these factors combined have resulted in the extensive use of SEC to isolate EVs from many different biological sources^[82-84]. The disadvantages of SEC include the requirement for upstream sample concentration before isolation and sample dilution resulting from repeated elution steps. In contrast, AEC separates EVs based on surface charge interactions between negatively charged EV membranes and positively charged stationary phases^[85,86]. AEC enables efficient EV recovery and simultaneous concentration, but its performance can be affected by the heterogeneity of EV surface charge and the presence of co-isolated charged contaminants, which may compromise purity.

Precipitation methods for isolating EVs involve increasing the concentration of EVs through intermolecular interactions that promote the aggregation of EVs. Precipitation methods for isolating EVs are classified into two broad categories: PP and AP. PP methods depend on lipophobicity of polymeric compounds to encapsulate EV membranes, allowing them to be retrieved through low-speed centrifugation^[87,88]. However, PP methods frequently co-precipitate lipoproteins and create difficulties in completely removing polymer contaminants from samples. AP methods encompass both immunoprecipitation (IP) and aptamer-based precipitation. IP utilizes antibody-functionalized magnetic beads or plates to selectively capture EV subpopulations via surface markers^[89,90]; although highly specific, it often suffers from limited elution efficiency. Conversely, aptamer-based precipitation enables reversible binding through conformational switching, facilitating gentler recovery of intact EVs^[91]. These methods are still limited in the throughput of EVs and are therefore primarily suited for laboratories performing smaller-scale or analytical applications.

In practical workflows, TFF is frequently combined with SEC^[92-94], UC^[95], or IP^[90] to improve EV purity^[96]. Additional strategies integrate AEC with SEC^[97] or UF^[98] to enhance separation efficiency. Despite these advances, no single or combined isolation strategy can simultaneously achieve optimal purity, specificity, recovery, and cost-effectiveness. Therefore, the selection of EV isolation approaches should be guided by the intended application, with careful consideration of process complexity, operational cost, and scalability.

High-resolution methodologies for characterization have made a significant impact on how we understand the function of MSC-EVs and their ability to move towards clinical application. As shown in Figure 2D (i), current methods for characterizing EVs use bulk methodology, such as enzyme-linked immunosorbent assay (ELISA), Western blotting (WB), and Dynamic Light Scattering (DLS), to measure properties such as the amount of protein in the total EV preparation, expression of specific markers, and the average size of the EVs^[99]. These methods continue to be a valuable method to assess the quality of EVs; however, these methods provide only population level data about EVs. All EV populations contain a high level of heterogeneity regarding the size, refractive index, and chemical makeup, resulting from the differences in cellular origin, physiological state, and biogenetic pathway^[100]. Bulk methodologies obscure particle-level variability and are easily contaminated by the presence of co-isolated non-vesicular components in the EV preparation^[101].

To address these limitations, single-particle methodologies have been increasingly developed as a means to obtain more precise measures of nanoscale properties and to better understand the heterogeneity of EVs [Figure 2D (ii)]. Nanoparticle Tracking Analysis (NTA) and Tunable Resistive Pulse Sensing (TRPS) are two of the most common methods used for measuring the size distribution and concentration of EVs. However, for both techniques, their low detection limits (typically 70-90 nm for NTA and 70-100 nm for TRPS) create challenges for successfully measuring sEVs^[102].

Electron microscopy (EM) techniques utilize electron beams with extremely short wavelengths in combination with high-precision optical systems and high accelerating voltages, enabling nanometer- and even subnanometer-scale resolution^[103]. Atomic force microscopy (AFM), transmission electron microscopy (TEM), scanning electron microscopy (SEM), and cryo-transmission electron microscopy (cryo-TEM) are widely applied to characterize EV size and morphology. AFM enables measurement of EV mechanical properties, SEM provides enhanced 3D surface visualization, and TEM offers superior spatial resolution. Nevertheless, these methods typically require extensive sample preparation steps - such as adhesion, fixation, dehydration, conductive coating, or negative staining - which may alter EV structure and morphology, thereby affecting analytical accuracy. By contrast, cryo-TEM preserves EVs in a near-native hydrated state by rapid vitrification in liquid nitrogen, allowing high-fidelity imaging of EV size distribution, surface morphology, purity, and internal structure. Despite these advantages, the complexity of sample preparation, long imaging times, and high equipment costs limit the widespread application of cryo-TEM. Distinct from EM-based approaches, super-resolution fluorescence microscopy (SRFM) overcomes the optical diffraction limit through fluorescence labeling and diverse imaging principles, achieving nanometer-scale resolution^[104]. For example, Wong-Dilworth et al. employed stimulated emission depletion (STED) microscopy to resolve the nanoscale localization of endogenously labeled ARF GTPases, elucidating their role in EV trafficking, while Jeong et al. used stochastic optical reconstruction microscopy (STORM) to visualize EV biogenesis in Gram-positive bacteria, revealing ultrastructural membrane dynamics^[105,106]. Nevertheless, SRFM techniques remain constrained by complex sample preparation, relatively low imaging throughput, and limited statistical representativeness, which collectively restrict broader application.

Flow cytometry (FCM) is a high-throughput, multiparametric analytical technique that enables single-particle detection and statistical analysis of particles in suspension. However, the lower detection limit of conventional FCM is typically above approximately 300 nm, rendering the light scattering signals of most sEVs indistinguishable from background noise^[107]. To overcome this limitation, nano-flow cytometry (nFCM) has been specifically optimized for detecting the weak light scattering signals of subwavelength nanoparticles. Combined with single-molecule fluorescence detection, nFCM enables highly sensitive single-particle analysis of EVs as small as 40 nm^[108]. The improved sensitivity is primarily achieved through optical and fluidic optimizations, including hydrodynamic focusing to reduce the optical detection volume, prolonging the particle transit time through the laser beam to generate detectable photon bursts, the use of high-quantum-efficiency avalanche photodiodes for weak scattering and fluorescence detection, and optimized optical filtering to suppress background stray light^[109]. Collectively, these optimizations improve scattering sensitivity by approximately 4-6 orders of magnitude and fluorescence sensitivity by approximately 2-3 orders of magnitude compared with conventional FCM^[110]. It should be noted that particle sizes measured by nFCM represent the optical equivalent diameter derived from Mie scattering theory rather than the true geometric diameter^[111]. The calculated size therefore depends on assumptions regarding the refractive index of EVs as well as the optical properties of the calibration particles. Current commercial platforms typically employ silica nanoparticles (n ≈ 1.43-1.46) as calibration standards and convert scattering intensity into particle size using a Mie scattering model assuming an EV refractive index of approximately 1.37-1.42, thereby reducing systematic errors arising from refractive index mismatches between calibration particles and EVs^[112]. Nevertheless, because the refractive index of EVs is intrinsically heterogeneous, the resulting particle sizes remain model-dependent estimates. Studies have shown that particle size distributions obtained by Mie-corrected nFCM are in good agreement with those measured by cryo-TEM, supporting the use of nFCM as a reliable optical approach for EV size estimation^[108,113]. In addition to light scattering, nFCM simultaneously detects fluorescence signals from individual EVs. By combining fluorescent antibodies^[108], nucleic acid dyes^[114], lipid membrane probes^[115], and sialic acid labeling strategies^[116], it enables multiparametric characterization of EV surface proteins, nucleic acids, membrane structures, and glycosylation profiles. It should be recognized, however, that fluorescence signals represent semi-quantitative molecular phenotypes based on fluorescence intensity. Their quantitative interpretation is influenced by antibody labeling efficiency, molecules of equivalent soluble fluorochrome (MESF) calibration, instrument sensitivity, and threshold effects associated with low-abundance targets^[117]. Consequently, fluorescence intensity measured from individual EVs should not be interpreted as a direct surrogate for molecular copy number and therefore requires appropriate calibration and experimental validation.

Beyond these considerations, several methodological limitations of nFCM should also be noted. First, similar to other light scattering-based single-particle techniques such as NTA, scattering signals alone are insufficient to reliably distinguish EVs from similarly sized non-vesicular particles, including lipoproteins and protein aggregates, in the absence of specific fluorescent labeling. Therefore, combining scattering measurements with EV-specific molecular markers is generally required to improve analytical specificity. Second, highly concentrated samples are prone to coincidence or swarm detection^[118], rendering the results sensitive to sample dilution, gating strategies, and data analysis workflows. Accordingly, standardized operating procedures (SOPs) are essential for improving experimental reproducibility and interlaboratory consistency. Furthermore, the lack of universally accepted reference materials and cross-platform calibration standards remains a major obstacle to the comparability of single-particle EV analyses across different instruments and laboratories^[119]. Despite these limitations, nFCM remains one of the few techniques capable of simultaneously measuring the scattering and fluorescence properties of individual EVs, making it an indispensable tool for investigating EV heterogeneity and functional subpopulations.

Overall, nFCM provides a powerful platform for multiparametric single-particle characterization of EVs. Nevertheless, its analytical performance remains constrained by assumptions inherent in optical scattering models, the semi-quantitative nature of fluorescence measurements, and the current lack of standardized analytical frameworks. Consequently, no single characterization technique is capable of comprehensively assessing EV size, concentration, morphology, molecular composition, and biological heterogeneity. Future studies should therefore integrate orthogonal characterization strategies - including bulk analyses, single-particle measurements, and high-resolution imaging - to cross-validate the physicochemical properties, molecular composition, and ultrastructure of EV preparations. Such complementary approaches will help minimize the intrinsic biases of individual techniques, improve the robustness and reproducibility of EV characterization, and provide a stronger methodological foundation for EV quality assessment, mechanistic studies, and clinical translation.

Building on this framework, the available evidence is discussed under three interconnected biological effects of EVs: preservation of RGC survival, modulation of neuroinflammation, and maintenance of metabolic homeostasis [Figure 3A]. Notably, current evidence suggests that axonal regeneration and visual recovery are more appropriately viewed as downstream therapeutic outcomes resulting from the combined actions of these protective mechanisms, rather than as independent mechanisms of EV-mediated neuroprotection [Figure 3B].

RGC apoptosis represents the final common pathological event in virtually all optic neuropathies and is currently the best-characterized therapeutic target of EV-based interventions. Accumulating evidence demonstrates that EVs derived from diverse cell types enhance RGC survival through multiple prosurvival signaling pathways, including the phosphoinositide 3-kinase (PI3K)/protein kinase B (AKT), cAMP response element-binding protein (CREB), c-Jun N-terminal kinase (JNK), and B-cell lymphoma 2 (Bcl-2) family signaling athways. For example, EVs released from hypoxia-preconditioned astrocytes, enriched in miR-329-5p, alleviate RGC apoptosis and retinal thinning after ischemia-reperfusion injury through inhibition of MAPK8/JNK signaling^[129]. Schwann cell-derived EVs are efficiently internalized by RGCs and promote RGC survival and axon regeneration through CREB signaling activation^[130]. In optic nerve crush (ONC) models, bone marrow MSC (BMSC)-derived EVs mediate anti-apoptotic effects through PI3K/AKT/Bcl-2 signaling, thereby reducing RGC apoptosis and preserving retinal integrity^[131]. Umbilical cord MSC-derived EVs enhance glial cell-mediated neurotrophic support, contributing to improved RGC survival^[132]. EVs derived from TNF-α-preconditioned gingival MSCs are enriched in miR-21-5p and improve RGC survival after ischemic injury through modulation of the miR-21-5p/PDCD4 axis and attenuation of inflammatory responses^[133]. Similarly, urine-derived stem cell EVs reduce apoptosis and cellular senescence during age-associated RGC degeneration^[134]. In glaucoma models, BMSC-derived sEVs exert neuroprotective effects by chronically preserving RGCs, with reported benefits in experimental glaucoma models^[135,136]. Collectively, these studies consistently demonstrate enhanced RGC survival and preservation of retinal structure, indicating that neuronal protection represents the most reproducible biological effect of EV-based therapy in optic neuropathies. However, RGC apoptosis does not occur in isolation. Persistent neuroinflammation amplifies neuronal injury through pro-inflammatory cytokine release, oxidative stress, and aberrant glial activation. Consequently, modulation of the neuroimmune microenvironment has emerged as another major mechanism underlying EV-mediated neuroprotection.

Persistent neuroinflammation is widely recognized as a major driver of secondary optic nerve degeneration. Increasing evidence indicates that EVs modulate the injured neuroimmune microenvironment by regulating communication among microglia, macrophages, and glial cells. Interferon gamma (IFN-γ)-preconditioned MSC-sEVs enriched in miR-423-5p reshape the retinal immune microenvironment by suppressing pro-inflammatory microglial activation while promoting protective phenotypes, resulting in greater neuroprotection than naïve MSC-derived EVs in ONC models^[137]. Astrocyte-derived EVs regulate glial responses and tissue homeostasis through the thrombospondin 1 (THBS1)/p21-activated kinase 3 (PAK3)/glutathione S-transferase mu 1 (Gstm1) signaling axis, thereby alleviating secondary injury^[138]. Human adipose-derived MSC-EVs attenuate microglial activation by suppressing Toll-like receptor 4 (TLR4)/mitogen-activated protein kinase (MAPK)/nuclear factor kappa B (NF-κB) signaling^[139]. In addition, functionally engineered EVs overexpressing miR424 (FEE-424) EVs delivering miR-424-5p further reduce ROS production and inflammatory cytokine expression, resulting in enhanced cytoprotection^[140]. Despite considerable variation in EV source and cargo composition, these interventions consistently attenuate inflammatory responses, suppress aberrant microglial activation, and restore tissue homeostasis, highlighting immunomodulation as a central component of EV-mediated neuroprotection. Importantly, neuroinflammation is closely intertwined with metabolic dysfunction. Mitochondrial impairment and disrupted energy metabolism not only compromise RGC survival directly but also exacerbate neuronal injury through excessive ROS production and activation of inflammatory signaling pathways. Consequently, increasing attention has been directed toward the role of EVs in regulating cellular metabolism and mitochondrial homeostasis.

Metabolic dysfunction, particularly mitochondrial impairment, oxidative stress, and NAD⁺ depletion, is increasingly recognized as a common pathological feature of glaucoma, TON, and inherited optic neuropathies. Compared with studies on anti-apoptotic and immunomodulatory mechanisms, investigations into EV-mediated metabolic regulation remain relatively limited. Nevertheless, growing evidence supports its contribution to neuroprotection. In an LHON-like model carrying the mtDNA m.11778G>A mutation, rat MSC-derived Super-EV-Mito improved visual function, increased ATP production, and reduced ROS accumulation, effects associated with mitochondrial transfer and restoration of oxidative phosphorylation^[141]. Nicotinamide-loaded adipose MSC-derived EVs improved NAD⁺ metabolism, thereby attenuating dendritic degeneration and preserving neuronal integrity^[142]. These findings indicate that, beyond conventional regulation of intracellular signaling pathways, EVs may protect injured neurons by restoring mitochondrial function, maintaining energy homeostasis, and enhancing cellular resilience under metabolic stress. However, current evidence is largely confined to individual disease models and candidate mechanisms, and the general contribution of metabolic regulation across different optic neuropathies remains to be established. Although promotion of neuronal survival, suppression of neuroinflammation, and maintenance of metabolic homeostasis target distinct pathological processes, they collectively improve the injured neural microenvironment and provide a favorable foundation for subsequent axonal repair.

Several studies have reported that EVs from different cellular sources promote axonal extension and improve visual function to varying degrees. For example, BMSC-derived EVs promote axonal regeneration and RGC survival through the delivery of exosomal miRNAs^[143]; hESC-MSC-derived EVs enhance axonal regeneration and long-term neuroprotection by suppressing cis-pTau accumulation^[144]; and huMSC-derived EVs facilitate axonal growth and preserve retinal nerve fiber layer thickness through exosomal miRNA-mediated activation of mechanistic target of rapamycin complex 1 (mTORC1) signaling^[145]. Notably, although these studies implicate different molecular pathways, axonal growth is generally accompanied by enhanced RGC survival, attenuation of neuroinflammation, or increased neurotrophic support, suggesting that the observed regenerative responses are likely secondary to improvements in neuronal viability and the local tissue microenvironment. From the perspective of optic nerve regeneration, however, the current evidence remains limited. Most studies evaluate increased axonal length, preservation of tissue structure, or short-term improvements in visual function, whereas convincing evidence demonstrating that regenerated axons can traverse the lesion, establish appropriate target connections, and restore functional visual circuits remains lacking. Furthermore, as part of the central nervous system, optic nerve regeneration is constrained not only by the intrinsic growth capacity of RGCs but also by glial scar formation, inhibitory extracellular cues, and impaired synaptic reconstruction. Taken together, the available evidence supports the conclusion that EVs primarily enhance the regenerative potential of injured axons by improving neuronal survival and remodeling the injury microenvironment, rather than directly driving long-distance functional optic nerve regeneration. Accordingly, axonal regeneration and visual recovery are better regarded as higher-order therapeutic outcomes resulting from the coordinated actions of multiple neuroprotective mechanisms, rather than as independent mechanisms of EV action. Future studies integrating axonal tracing, electrophysiological assessment, and behavioral analyses will be essential for defining the true contribution of EVs to functional neural circuit reconstruction and optic nerve regeneration.

Although a growing body of evidence supports the therapeutic potential of EVs in optic neuropathies, a critical appraisal of the studies summarized in Table 2 reveals substantial methodological heterogeneity and variability in evidence quality, limiting both cross-study comparisons and the robustness of therapeutic conclusions. First, considerable heterogeneity exists in EV isolation and characterization methods. Different studies employ UC, UF, PEG precipitation, or commercial isolation kits, while quality control criteria remain inconsistent and adherence to the MISEV guidelines is often incomplete. Second, EV dosing lacks standardization. Dosages are variously reported as particle number, protein content, or administration volume, making quantitative comparisons and effect size analyses difficult. Third, experimental controls are frequently inadequate. Most studies include only phosphate-buffered saline controls, whereas critical controls such as EV-depleted conditioned medium, heat-inactivated EVs, or process blanks are often absent, making it difficult to exclude the contribution of non-vesicular components. Fourth, reporting of potential sources of bias remains insufficient. Key methodological details - including randomization, blinded outcome assessment, sample size calculation, and effect size estimation - are frequently omitted, compromising the overall quality of the evidence. Finally, mechanistic validation remains relatively superficial. Most studies focus on candidate miRNAs or individual signaling pathways, whereas rigorous evidence establishing causality, such as cargo depletion, rescue experiments, and multi-omics integration, is still limited.

Beyond methodological standardization, a more fundamental biological question concerns how EV composition is linked to therapeutic efficacy. Although EVs derived from different sources differ substantially in their molecular cargo and proposed mechanisms of action, their therapeutic effects consistently converge on a limited number of shared pathological processes, including enhanced RGC survival, attenuation of neuroinflammation, and restoration of metabolic homeostasis. This coexistence of compositional heterogeneity and functional convergence suggests that the therapeutic actions of EVs cannot be fully explained by individual cargo molecules or isolated signaling pathways alone. Instead, current evidence indicates that distinct EV populations may achieve comparable neuroprotective outcomes by targeting the same or interconnected pathological nodes. Consequently, rather than continuously identifying new candidate cargos, future studies should focus on elucidating how diverse EVs modulate these shared disease processes. In other words, the true convergence lies not in EV composition itself, but in the pathological pathways they regulate, including RGC apoptosis, persistent neuroinflammation, and metabolic dysregulation.

This perspective naturally raises another important question: if EVs from distinct cellular sources achieve similar therapeutic outcomes through different molecular mechanisms, does this functional convergence arise from a common set of bioactive components, or from the coordinated actions of multiple functional EV subpopulations? Increasing evidence suggests that even EVs released by the same cell type exhibit substantial heterogeneity in surface markers, cargo composition, cellular uptake preference, and biological activity^[146]. Consequently, the therapeutic effects of EV preparations, which are typically evaluated as bulk populations, are likely to result from the collective actions of multiple functional subpopulations rather than from a single dominant cargo. Addressing this question will require establishing direct links between EV subpopulations, functional cargos, pathological targets, and therapeutic outcomes. Advances in single-particle analysis, high-dimensional phenotyping, and functional subpopulation isolation are expected to facilitate this transition from population-level characterization toward function-oriented analyses of discrete EV subpopulations. Such advances may also reshape EV quality assessment. Current quality control primarily relies on physicochemical parameters, including particle size distribution, particle concentration, and canonical marker expression, which largely reflect EV composition rather than biological activity. As functional EV subpopulations and key effector cargos become better defined, EV quality assessment is expected to evolve from conventional physicochemical characterization toward potency assays based on biological activity. Incorporating disease-relevant functional endpoints, such as RGC protection, immunomodulation, and maintenance of metabolic homeostasis, into quality control frameworks will provide a more reliable basis for developing function-enhanced EV products, optimizing manufacturing processes, and ensuring consistent clinical quality.

Nevertheless, generating high-quality EV preparations alone is unlikely to ensure therapeutic efficacy. Efficient delivery to target tissues and consistent therapeutic performance remain equally critical for successful clinical translation. At present, intravitreal injection is the predominant route of EV administration in optic neuropathy studies because it enables direct delivery to the retina and RGCs. However, this approach is invasive, often requires repeated administration, and carries risks of infection, hemorrhage, and retinal injury^[147]. In contrast, intravenous and topical ocular administration are clinically more convenient but generally suffer from limited delivery efficiency owing to the blood-retinal barrier and rapid ocular clearance^[148]. More recently, intranasal administration has emerged as a promising non-invasive delivery strategy. Experimental studies have demonstrated that EVs can access the central nervous system through olfactory and trigeminal nerve-associated pathways, thereby partially bypassing the blood-brain barrier^[149,150]. Compared with repeated intraocular injections, intranasal delivery offers advantages including ease of administration, improved patient compliance, and suitability for long-term treatment. Although studies evaluating intranasal EV delivery in optic neuropathies remain limited, encouraging results in other central nervous system disorders suggest that this strategy warrants further investigation as a clinically feasible long-term therapeutic approach for chronic optic neuropathies such as glaucoma.

Meanwhile, the development of engineered EVs is providing new opportunities to enhance therapeutic efficacy while addressing the intrinsic heterogeneity of native EVs. Various engineering strategies - including donor cell preconditioning, genetic modification, and functional cargo loading - have been employed to generate EVs with enhanced therapeutic properties^[151]. For example, inflammatory cytokine preconditioning enriches MSC-derived EVs with immunoregulatory miRNAs, thereby augmenting their anti-inflammatory and neuroprotective activities^{[133,152,153]}. Likewise, loading EVs with functional cargos such as miRNAs, neurotrophic factors, nicotinamide, or mitochondria-related components further enhances their ability to target specific pathological processes^[154]. In addition, membrane engineering and ligand-mediated surface modification can improve EV targeting toward RGCs and injured optic nerve tissues^[155]. More importantly, engineered EVs can be rationally designed to target specific pathological nodes, thereby enhancing their capacity to promote RGC survival, regulate neuroinflammation, or restore metabolic homeostasis. Compared with native EVs, engineered EVs also offer the potential for more consistent quality control, better-defined mechanisms of action, and improved therapeutic reproducibility, making them a promising direction for advancing EV-based therapeutics toward clinical application.

Overall, future progress in EV-based therapy will depend not only on continued optimization of EV sources and manufacturing processes but also on a deeper understanding of the relationships among functional EV subpopulations, key effector cargos, and shared pathological nodes. Integrating standardized manufacturing, high-resolution quality assessment, targeted delivery strategies, and EV engineering is expected to facilitate the transition of EV therapeutics from empirical applications toward mechanism-driven, precision interventions, thereby providing a stronger scientific and technological foundation for the clinical translation of EV-based therapies for optic neuropathies.