Alzheimer’s disease (AD) is a common neurodegenerative disorder in older adults, characterized by progressive cognitive decline and behavioral disturbances associated with degeneration of the central nervous system (CNS)^[1,2]. As the most common form of dementia, AD accounts for approximately 60%-70% of all dementia cases, and the global number of people living with dementia is projected to exceed 139 million by 2050^[3,4]. This rapidly increasing disease burden poses substantial challenges to healthcare systems and the global economy. The hallmark neuropathological features of AD include extracellular amyloid-β (Aβ) plaques, intracellular neurofibrillary tangles (NFTs) composed of hyperphosphorylated tau, neuronal loss, synaptic degeneration, granulovacuolar degeneration, and gliosis^[5-7]. The accumulation of pathogenic Aβ can activate microglia to trigger chronic neuroinflammatory responses, thereby exacerbating tau phosphorylation and neuronal injury^[8]. These interconnected pathological events may form a self-reinforcing cycle that links amyloid pathology, tau pathology, neuroinflammation, and progressive neurodegeneration^[9]. A major obstacle to the development of effective AD therapeutics is the presence of the blood-brain barrier (BBB). Although the BBB plays a key role in protecting the brain by restricting the entry of harmful substances from the bloodstream, it also limits the penetration of many therapeutic agents, including both small-molecule drugs and biologics^[10]. Many AD drug candidates, especially Aβ-targeting antibodies, have poor BBB permeability due to their size, polarity, or stability. As a result, they fail to reach effective concentrations in the brain, which may partly explain their limited success in clinical trials despite promising early results.

Extracellular vesicles (EVs) are nanoscale lipid bilayer particles secreted by most cell types and have emerged as a promising platform for disease modulation and targeted drug delivery in AD^[11]. Due to their favorable biocompatibility, low immunogenicity, and natural ability to cross the BBB, EVs offer clear advantages as delivery vehicles^[12]. They can carry various bioactive molecules, including nucleic acids, proteins, lipids, and metabolites, directly to recipient cells, enabling precise regulation of disease-related processes^[13,14]. To improve consistency in EV research, the International Society for Extracellular Vesicles (ISEV) introduced the minimal information for studies of extracellular vesicles (MISEV2023) guidelines, which standardize EV classification, isolation, characterization and reporting, thereby enhancing reproducibility and clinical translation^[15,16]. Stem cell-derived EVs (SC-EVs), particularly those derived from mesenchymal stem cells (MSCs), neural stem cells (NSCs), and induced pluripotent stem cells (iPSCs), have shown multiple therapeutic effects in AD models. These include reducing Aβ production, limiting tau phosphorylation, promoting synaptic plasticity, enhancing neurogenesis, and suppressing neuroinflammation and oxidative stress^[17-20]. Notably, these effects partly reflect the functions of their parent cells while avoiding risks such as immune rejection and tumor formation. Recent advances in EV engineering strategies, including surface functionalization, cargo loading, stimulus-responsive release, and improved delivery methods such as intranasal administration, have further increased their brain-targeting ability and therapeutic potential. In addition to mammalian EVs, plant-derived exosome-like nanovesicles (PELNVs) are gaining attention as scalable and orally administered delivery platforms. Their potential to regulate the gut-brain axis (GBA) may provide added benefits for CNS intervention. Following the MISEV2023 guidelines, this review uses “extracellular vesicles (EVs)” as a general term for cell-derived lipid bilayer particles. The term “EVs” is used consistently, while “exosomes” is applied only when specifically defined in the cited studies.

EVs can regulate multiple pathological processes and also serve as delivery platforms, making them a promising strategy for next-generation AD therapies. However, AD is not caused by a single factor; rather, it results from a complex, interconnected network of pathological events, including Aβ accumulation, tau pathology, neuroinflammation, oxidative stress, mitochondrial dysfunction, synaptic impairment, reduced neurogenesis, and GBA imbalance. Therefore, EVs should not be regarded simply as passive carriers. Instead, they act as mediators of intercellular communication that can either spread signals or help restore biological balance. These effects depend on their cellular source, molecular cargo, disease stage, and engineering design. This review examines EV mechanisms of action, engineering approaches, delivery strategies, manufacturing requirements, safety considerations, and clinical translation potential in AD treatment.

AD is not driven by a single pathological process, but rather by a self-reinforcing network involving Aβ accumulation, tau pathology, neuroinflammation, oxidative stress, mitochondrial dysfunction, synaptic impairment, reduced neurogenesis, and GBA imbalance^[21,22]. These processes are closely interconnected. Aβ accumulation can activate microglia and astrocytes, whereas persistent neuroinflammation can promote tau phosphorylation and synaptic damage^[23,24]. Mitochondrial dysfunction increases oxidative stress, while redox imbalance further accelerates protein aggregation and neuronal degeneration^[25,26]. Downstream changes, including synaptic loss, demyelination, cholinergic dysfunction, and impaired hippocampal neurogenesis, directly contribute to cognitive decline^[27-30]. Within this network, EVs should not be viewed as simple carriers acting on single targets. Instead, they act as mediators of intercellular communication, transferring proteins, lipids, nucleic acids, inflammatory factors, mitochondrial regulators, and neurotrophic signals among neurons, glial cells, immune cells, and GBA-related pathways. Depending on their origin, cargo, and disease context, EVs can either spread pathological Aβ and tau species and inflammatory signals or support Aβ clearance, immune regulation, mitochondrial stability, and neural repair. Their therapeutic potential in AD derives from this ability to reshape the pathological network, shifting it away from a self-reinforcing degenerative state toward clearance, repair, and homeostasis.

AD is a multifactorial neurodegenerative disease. Therefore, EV-based therapies are unlikely to fully address its pathological complexity through a single intervention. A more clinically realistic strategy may involve combining EV-based approaches with approved symptomatic treatments, disease-modifying antibodies, or lifestyle interventions. Such combination strategies could more precisely target the multiple pathological features of AD by synchronously modulating cognitive symptoms, amyloid pathology, neuroinflammation, mitochondrial dysfunction, and neural repair processes, thereby better aligning with the interconnected nature of the disease.

One feasible strategy is to combine EV-based therapy with AChEIs, such as donepezil, rivastigmine, and galantamine. As mentioned above, AChEIs mainly relieve symptoms by enhancing cholinergic neurotransmission^[116]. EVs can exert non-cholinergic regulatory effects by reducing neuroinflammation, oxidative stress, mitochondrial dysfunction, neuronal apoptosis, and synaptic injury^[117]. This mechanistic complementarity suggests that the combination of EVs and AChEIs can simultaneously target cognitive symptoms and underlying cellular damage in AD. Such combinations may provide symptomatic cholinergic support together with broader neuroprotective effects, especially for patients with early or mild-to-moderate AD. Further research is still needed to clarify the optimal dose, the timing of administration, the interaction between drugs and EV-based therapy, and long-term safety. Another promising strategy is to combine EV-based therapy with anti-Aβ antibodies. Anti-Aβ antibodies mainly reduce amyloid plaque burden, while EVs can regulate downstream pathological processes, including microglial activation, tau phosphorylation, mitochondrial dysfunction, oxidative stress, and synaptic degeneration^[41,44,46]. Engineered EVs carrying anti-inflammatory molecules, miRNAs, siRNAs, neurotrophic factors, mitochondrial regulators, or small-molecule drugs can complement antibody-mediated amyloid clearance and provide broader neuroprotective effects^[118-120]. Against this background, EVs may help shift the pathological network of AD from amyloid-driven inflammation and neuronal damage to immunomodulation, mitochondrial stabilization, and neural repair. However, such combination therapies need to be carefully evaluated, with particular attention to factors such as immune activation, vascular effects, amyloid-related imaging abnormalities, treatment timing, and patient selection.

EV-based therapy may also be combined with lifestyle interventions, such as dietary modification, physical exercise, and metabolic regulation. These interventions can influence systemic inflammation, gut microbiota, oxidative stress, insulin sensitivity, neurovascular function, and synaptic plasticity, all of which are closely associated with AD progression^[121]. EVs may participate in these processes by mediating communication between the periphery and the CNS. For example, exercise- or diet-induced changes in circulating EV cargo may reflect or contribute to alterations in neuroinflammation, mitochondrial function, and neuronal repair^[122,123]. Thus, EVs may also serve as biomarkers for monitoring physiological responses to lifestyle interventions. Combining EV-based therapies with exercise, dietary strategies, or metabolic interventions may provide a more personalized and multidimensional approach to AD management. EV-based combination therapy may better align with the multifactorial and interconnected nature of AD than single-target treatments. By integrating cholinergic symptom control, amyloid clearance, anti-inflammatory effects, mitochondrial support, and neural repair, such strategies may expand the therapeutic window of EV-based AD therapy. However, most of these approaches remain at the conceptual or preclinical stage. Further mechanistic, pharmacological, and clinical studies are needed to evaluate the therapeutic synergy of EV-based combination strategies and define optimal dosing regimens, treatment timing, safety profiles, biomarker responses, and the patient populations most likely to benefit.

The translational potential of EVs as therapeutic carriers and bioactive agents for AD largely depends on the stability, purity, batch-to-batch consistency, and scalability of their production processes^[124]. According to the MISEV2023 guidelines, EV studies should systematically report sample sources, preprocessing conditions, isolation and concentration methods, EV characterization parameters, and functional validation strategies to improve reproducibility and comparability across studies^[16]. Evaluation of EVs for AD therapy should consider both their biological functions, including neuroprotection, anti-inflammatory effects, inhibition of Aβ deposition, and modulation of tau pathology, and the extent to which their isolation, purification, and quality control processes meet the standards required for clinical translation.

At present, the main EV isolation methods include differential ultracentrifugation, density gradient centrifugation, ultrafiltration, size-exclusion chromatography (SEC), immunoaffinity capture, and microfluidic technologies^[125]. A comparison of commonly used EV isolation techniques is presented in Table 2. Differential ultracentrifugation is the most classical and widely used method and remains suitable for basic research. However, it is time-consuming, exhibits considerable variability in recovery efficiency, and may result in the co-precipitation of protein aggregates, lipoproteins, or other non-EV components^[126]. Density gradient centrifugation can improve EV purity, but the procedure is technically complex, has limited yield, and is difficult to adapt to large-scale production^[124]. SEC offers several advantages, including mild processing conditions and relatively good preservation of EV integrity and biological activity, and has therefore been increasingly applied to the purification of therapeutic EVs in recent years. However, its sample-processing capacity remains limited, and sample dilution may occur during separation^[127]. In contrast, tangential flow filtration (TFF) provides better scalability, is suitable for processing large volumes of cell culture supernatants or biological fluid samples, and can be combined with SEC to balance concentration efficiency, purity, and preservation of biological activity^[128]. In recent years, microfluidic platforms have received increasing attention as alternatives to traditional EV isolation methods. These systems integrate EV capture, purification, detection, and analysis into microdevices and offer several advantages, including low sample consumption, rapid processing, high automation, and precise separation of specific EV subpopulations^[129]. Immunoaffinity capture enables the selective isolation of EVs based on specific surface markers or defined EV subpopulations. This method offers high specificity and is particularly useful for diagnostic applications, biomarker studies, and the enrichment of specific EV subgroups, such as brain-derived or neuron-derived EVs. However, immunoaffinity capture is limited by high cost and relatively low processing capacity, and the selection of specific markers may introduce bias by enriching only marker-positive EV populations^[130,131]. In AD research, microfluidic technologies are valuable for the detection and analysis of brain-derived or neuron-derived EVs and downstream biomarker studies^[132]. Nevertheless, several challenges remain, including limited standardization, variability in recovery efficiency due to differences in chip materials and antibody selection, interference from complex clinical samples, and difficulties in scaling up production^[133]. Therefore, microfluidic technologies are currently more suitable for EV-based diagnostics and subpopulation analysis, whereas large-scale production of therapeutic EVs still requires integration with manufacturing processes that are easier to scale up.

Quality control of therapeutic EVs is a core component of their clinical translation. An ideal EV formulation should be evaluated across multiple dimensions, including identity confirmation, purity, safety, potency, and stability. Common characterization methods include nanoparticle tracking analysis (NTA) or tunable resistive pulse sensing (TRPS) for assessing particle size distribution and concentration; transmission electron microscopy (TEM) or cryo-electron microscopy (cryo-EM) for observing vesicle morphology; and Western blotting, flow cytometry, or enzyme-linked immunosorbent assay (ELISA) to detect EV-enriched markers such as CD9, CD63, CD81, TSG101, and ALIX^[134-136]. In parallel, non-EV contaminants or negative markers should also be assessed, together with potential contamination by free proteins or nucleic acids. For engineered or drug-loaded EVs, additional evaluations are required, including drug or nucleic acid loading efficiency, release kinetics, targeting capability, residual transfection reagents, residual host-cell DNA, sterility, mycoplasma contamination, and endotoxin levels^[124]. In the context of AD therapy, EV quality control should also be integrated with disease-relevant functional assays. For example, the potency of EV formulations can be evaluated using assays related to Aβ aggregation or clearance, tau phosphorylation, microglial inflammatory responses, oxidative stress, neuronal survival, synaptic protein expression, and BBB penetration. Compared with assessments based solely on particle size, concentration, and surface markers, potency assays directly linked to AD-related pathological mechanisms are more informative for determining whether different EV batches retain consistent therapeutic activity. This consideration is important for MSC-EVs, NSC-EVs, engineered EVs, and PELNVs alike.

Scalable manufacturing is a key issue that must be addressed before EV-based therapies can advance toward clinical application. Current promising strategies include the use of serum-free or xeno-free culture systems, establishment of standardized donor cell banks, the adoption of 3D culture systems or bioreactors to increase EV yield, and downstream concentration and purification using TFF, SEC, or chromatography-based platforms^[137]. For clinical-grade EV products, closed manufacturing processes under good manufacturing practice (GMP) conditions must also be established, with clearly defined control of raw materials, batch production records, release criteria, storage conditions, and transportation stability. Recent studies have emphasized that EV drug development requires not only demonstration of biological function but also the establishment of traceable, reproducible, and verifiable manufacturing and quality control systems^[124]. Overall, EV isolation and purification technologies are gradually shifting from traditional laboratory-scale ultracentrifugation methods toward TFF, SEC, microfluidic systems, automation, and GMP-compatible platforms. For AD therapy, the key future challenge is not only to demonstrate the neuroprotective effects of EVs but also to establish a complete technical pipeline covering cell source, culture conditions, isolation and purification, quality control, potency validation, and scalable manufacturing. Only after the identity, purity, safety, stability, and therapeutic potency of EV formulations have been systematically validated can EV-based AD therapy progress from experimental research toward clinical translation.

In addition to their therapeutic potential, EVs are increasingly recognized as non-invasive diagnostic and early-stage biomarkers for AD. EVs isolated from biofluids such as blood, cerebrospinal fluid (CSF), saliva, or urine may carry disease-relevant cargo^[32]. Because EVs partly reflect the molecular status of their parental cells, neuron-derived or brain-enriched EV subpopulations may provide valuable information for early diagnosis, disease staging, and pathological monitoring. EV-based biomarkers may also facilitate patient stratification, treatment response monitoring, target engagement assessment, and the identification of individuals most likely to benefit from EV-based or antibody-based therapies. Therefore, EV diagnostics may complement EV therapeutics and support a more personalized translational framework for AD management.

Engineering strategies are central to improving the therapeutic performance of EVs in AD, as native EVs often exhibit insufficient brain targeting, heterogeneous cargo composition, variable therapeutic potency, and limited delivery efficiency^[138]. However, EV engineering should not be regarded as a single technical category. Instead, different strategies act at distinct stages of EV development, including donor-cell manipulation, surface modification, therapeutic cargo loading, hybrid or biomimetic platform construction, stimuli-responsive release, and delivery-route optimization. Therefore, a hierarchical framework is needed to distinguish engineering methods, representative platforms, advantages, limitations, and translational considerations. In this review, EV engineering strategies for AD therapy are organized into six major levels: source-cell engineering, surface engineering, cargo engineering, hybrid or biomimetic EV platforms, stimuli-responsive EV systems, and delivery-route optimization. Source-cell engineering determines the intrinsic molecular cargo and biological activity of EVs. Surface engineering improves BBB penetration, brain targeting, cellular uptake, and circulation stability. Cargo engineering introduces therapeutic molecules that regulate Aβ pathology, tau pathology, neuroinflammation, oxidative stress, mitochondrial dysfunction, and neural repair. Hybrid or biomimetic systems combine EVs with synthetic platforms or cell membrane-based systems to enhance stability and targeting specificity. Stimuli-responsive systems allow controlled release of cargo at specific sites or times. Optimizing delivery routes can affect CNS exposure, systemic distribution, patient adherence, and the feasibility of clinical translation.

Efficient and safe delivery of EV-based therapeutics to the brain remains a key determinant of their translational feasibility in AD therapy. The route of administration significantly affects EV biodistribution, brain exposure, systemic clearance, immune responses, dosing frequency, patient compliance, and long-term safety. Therefore, selection of an appropriate administration route should not only maximize CNS accumulation but also to balance invasiveness, scalability, repeatability, and clinical practicality. Intranasal administration has emerged as one of the most attractive delivery routes for EV-based AD therapy^[165]. Preclinical and early clinical studies are increasingly using intranasal administration to deliver MSC-EVs or engineered EVs because this route may enhance CNS bioavailability while maintaining a favorable safety profile^[166]. However, intranasal delivery also has several limitations, including mucociliary clearance, enzymatic degradation in the nasal cavity, limited administration volume, interindividual anatomical variability of the nasal cavity, and variability in dose retention. Therefore, dose consistency, formulation, stability, nasal residence time, and long-term local safety should be carefully evaluated before clinical application. A structured comparison of these strategies is presented in Table 4 and Figure 3.

Intravenous injection is another widely used route of administration because it is technically straightforward, clinically well established, and suitable for large-scale systemic delivery. This route is particularly useful when EVs are designed to exert both central and peripheral immunomodulatory effects^[167]. However, intravenously administered EVs are often rapidly cleared by the mononuclear phagocyte system and tend to accumulate in organs such as the liver, spleen, lung, and kidney, resulting in limited brain delivery efficiency^[168]. Surface engineering strategies, including RVG peptide modification, ligand conjugation, and membrane functionalization, can improve BBB penetration and brain targeting. These modifications may also change the pharmacokinetic characteristics, immunogenicity, and toxicity of EVs. Therefore, intravenous EV therapy requires comprehensive biodistribution analysis, dose optimization, repeated-dose toxicity evaluation, and off-target organ accumulation evaluation.

Oral administration is particularly suitable for PELNVs. These vesicles have received wide attention because of their high natural abundance, low production cost, good biocompatibility, and suitability for large-scale production^[175]. Compared with mammalian cell-derived EVs, PELNVs may be more stable in the gastrointestinal environment and can interact with intestinal epithelial cells, immune cells, and gut microbiota^[175,176]. This characteristic makes oral administration particularly promising for GBA-targeted AD interventions. By regulating intestinal inflammation, microbial composition, metabolic signaling, and systemic immune response, oral PELNVs can indirectly affect CNS inflammation and neurodegenerative processes^[169]. However, oral EV administration also faces a number of major challenges, including gastrointestinal degradation, variable absorption efficiency, uncertain pharmacokinetic characteristics, inter-batch heterogeneity, and limited understanding of the mechanisms by which these EVs and their cargo reach or influence the brain. Future research should clarify their intestinal uptake mechanism, metabolic fate, bioactive cargo stability, and long-term safety.

Subcutaneous and intramuscular injections can be used as alternative routes for repeated EV administration, especially when continuous systemic exposure or immune modulation is required. Compared with intracerebral injection, these administration routes are less invasive; for long-term treatment, they are also more convenient than repeated intravenous infusion. They may also promote the gradual absorption of EVs through local tissue and lymphatic circulation, thus prolonging the circulation time and reducing the toxicity associated with peak concentration^[170]. However, whether subcutaneously or intramuscularly administered EVs can achieve sufficient CNS exposure remains uncertain. Local immune response, inflammation at the injection site, degradation by tissue resident phagocytes, and different lymphatic transport pathways may all limit therapeutic efficacy. At present, these administration routes are more suitable for exploratory research or systemic immunomodulatory strategies than for EV therapy directly targeting the brain.

Intracerebroventricular or intracerebral administration can directly deliver EVs to the CSF or brain parenchyma, thus increasing CNS exposure and reducing reliance on BBB transport^[171]. This method is suitable for mechanistic research and proof-of-concept experiments. Its clinical application in AD is limited by high invasiveness, risks of infection or bleeding, low patient acceptance, and poor suitability for repeated long-term administration. Direct CNS administration can achieve high delivery efficiency, but it is unlikely to become the preferred route for routine EV-based AD therapy.

Intrathecal administration is another potential route for delivering EV-based therapeutics to the CNS. Unlike intravenous administration, intrathecal injection introduces EVs directly into the CSF through the spinal subarachnoid space, bypassing the BBB and reducing systemic exposure^[173]. Compared with intracerebroventricular or intracerebral injection, this method can avoid direct cranial surgery and may have higher clinical feasibility in specific neurological diseases. Intrathecal administration is especially important for EVs that carry nucleic acids, proteins, or other macromolecular cargo that need to enter the CNS efficiently. Evidence supporting this route in AD remains limited. A recent systematic review of EV-based therapy pointed out that most in vivo studies use intravenous administration, followed by intranasal administration and intracerebral administration; intrathecal administration has not yet become the main administration strategy for Alzheimer’s disease-specific EV research^[172]. The potential value of intrathecal EV administration in AD is currently mainly based on broader CNS disease models and EV-mediated gene delivery research. In these studies, intrathecal delivery of EVs has been used to achieve CNS-directed nucleic acid delivery and has been verified in spinal cord injury models. The results show that EVs can reach the CSF compartment and may produce neuroprotective or regenerative effects^[174,177]. Intrathecal administration has potential advantages, but it is still an invasive procedure and there are some challenges in clinical translation and application. These risks include complications associated with the operation, such as infection, bleeding, meningeal irritation, uneven distribution of CSF, rapid CSF clearance, and limited feasibility for repeated administration. AD-specific studies should compare the effects of intrathecal administration with those of intranasal, intravenous, and intracerebral administration routes, and focus on drug distribution in the CSF, penetration from the CSF into brain parenchyma, dose-response relationship, safety characteristics, and cognitive function changes.

For EV-based AD treatment, there is no universally applicable single administration route. Intranasal delivery is suitable for non-invasive brain-targeted treatment, while intravenous injection is suitable for large-scale systemic administration. Oral administration has shown potential for PELNV-based GBA modulation, while subcutaneous or intramuscular injection may help achieve long-term systemic immunomodulation. Translational studies should compare these routes using standardized pharmacokinetic, biodistribution, safety, and efficacy endpoints. The optimal route depends on EV source, engineering strategy, cargo type, disease stage, treatment frequency, and whether the goal is direct CNS delivery or systemic modulation of AD-related pathology. EV engineering also supports combination therapy. By enhancing brain targeting, cargo loading, and controlled release, engineered EVs can be combined with existing AD treatments, including AChEIs, anti-Aβ antibodies, and non-pharmacological interventions, to achieve multi-target regulation of AD pathology.

EVs, particularly MSC-EVs, are promising candidates for AD therapy owing to their low immunogenicity, potential scalability from standardized donor cell lines, relatively non-invasive administration options, and favorable storage properties. Based on the literature identified from PubMed and ClinicalTrials.gov, with the last search conducted on May 30, 2026, clinical translation of EV-based AD therapies remains at an early stage but shows encouraging potential. Among the limited clinical studies identified, MSC-derived EVs and MSC-derived EV-containing preparations are the main therapeutic modalities reported to date, with intranasal administration explored for its ability to bypass the BBB and enable repeated, non-invasive dosing. The most advanced clinical example is the Phase I/II open-label study at Ruijin Hospital (NCT04388982)^[178], in which intranasal ahaMSC-EVs were given to patients with mild-to-moderate AD. The treatment showed a favorable safety profile, with no EV-related adverse events and early signs of cognitive improvement. A later peer-reviewed report supported these findings and provided dose references and feasibility data for future randomized controlled trials. A related single-arm feasibility study in Japan evaluated intranasal administration of MSC secretome containing EVs and soluble factors, rather than purified EVs, in patients with AD^[179]. Although the study did not use purified EVs, the intervention was well tolerated, improved Hasegawa Dementia Scale-Revised (HDS-R) scores, and showed that self-administration at home was feasible. These findings highlight the broader therapeutic potential of EV-rich biological agents and the need to clarify product definitions and standardization. Early clinical data are encouraging, but EV-based AD therapy still faces major translational challenges, including GMP-compatible large-scale manufacturing, standardized isolation and purification, reproducible batch release standards, dose selection for different administration routes, and disease-related efficacy evaluation. Future clinical trials should adopt a systematic treatment framework to evaluate the impact of EV interventions on cognitive function, Aβ/tau biomarkers, neuroinflammation, mitochondrial dysfunction, synaptic repair, functional outcomes, and long-term safety. Clinical studies of EV-based or EV-containing approaches in AD are summarized in Table 5.

Although EV-based therapies have demonstrated considerable potential in AD intervention, their clinical application necessitates the development of a specialized, systematic, and mechanism-informed safety assessment framework. Unlike traditional small-molecule drugs, EVs are biologically active, heterogeneous nanoscale particles, and their safety depends not only on the therapeutic cargo but also on multiple additional factors. These include the origin and physiological status of donor cells, culture and conditioning protocols, methods of EV isolation and purification, surface composition, engineering strategy, administration route, dosing frequency, disease stage, and batch-to-batch production consistency.

The origin of donor cells and the composition of EV cargo are fundamental determinants of EV safety. EVs derived from healthy or therapeutically primed MSCs, NSCs, astrocytes, or anti-inflammatory microglia may carry neuroprotective proteins, regulatory miRNAs, antioxidant molecules, and immunomodulatory signals^[180,181]. In contrast, EVs released from stressed neurons, disease-related glial cells, inflammatory donor cell cultures, or poorly controlled production systems may contain pathogenic Aβ species, phosphorylated tau proteins, pro-inflammatory cytokines, complement-related proteins, or dysregulated miRNAs^[180,182]. Such harmful cargo may promote abnormal protein aggregation, glial cell activation, synaptic impairment, and neuronal dysfunction. Therefore, EV preparations intended for AD therapy require comprehensive molecular characterization before clinical application. This should include assessment of Aβ/tau burden, inflammatory mediators, miRNA signatures, protein contaminants, and disease-related molecular patterns.

Immunogenicity and abnormal immune regulation should also be systematically evaluated. EVs, particularly MSC-derived EVs, are generally considered to exhibit relatively low immunogenicity^[183]. However, low immunogenicity does not imply the absence of immune-related risks. This problem is particularly relevant in AD because neuroinflammation is a central driver of disease progression^[60]. EVs engineered to suppress neuroinflammation may provide therapeutic benefits; nevertheless, excessive or non-specific immunomodulation may disrupt the surveillance function of microglia, the supportive and homeostatic roles of astrocytes, peripheral immune balance, or host defense mechanisms. Therefore, safety evaluation should include analyses of cytokine release, complement activation, anti-EV antibody formation, microglial activation state, astrocytic reactivity, and systemic immune profiles.

Biodistribution, pharmacokinetics, and off-target accumulation should be evaluated in a route-specific manner. Intravenous administration of EVs is clinically convenient and scalable; however, intravenously administered EVs are often rapidly taken up by the liver, spleen, lung, kidney, and mononuclear phagocyte system, which may reduce delivery efficiency and increase peripheral tissue exposure^[168]. Intranasal administration can enhance direct transport from the nasal cavity to the brain and reduce systemic exposure, but dose retention, mucociliary clearance, nasal epithelial safety, and interindividual variability remain important considerations. Oral delivery, particularly for PELNVs, may be suitable for modulating the GBA^[161]. However, the gastrointestinal degradation, absorption efficiency, metabolic fate, and CNS-related effects of orally administered EVs remain incompletely understood. Although direct CNS administration, including intrathecal, intracerebroventricular, or intracerebral delivery, can improve EV distribution in the CSF or brain parenchyma, it also introduces procedure-related risks, such as infection, hemorrhage, meningeal irritation, local tissue damage, and limited feasibility of long-term repeated administration. Therefore, route-specific studies of biodistribution, clearance kinetics, organ accumulation, and tolerability are essential for optimizing EV-based AD therapy.

Mechanism-specific CNS safety risks are closely related to the pathological processes targeted by EV-based interventions. EVs designed to modulate Aβ and tau pathology should be carefully screened to prevent the unintended transfer of pathogenic Aβ oligomers, tau seeds, or other aggregation-prone proteins. EVs developed to regulate neuroinflammation should preserve the physiological functions of microglia and astrocytes, including synaptic pruning, neurotransmitter recycling, BBB maintenance, and immune surveillance. EVs targeting oxidative stress or mitochondrial dysfunction should avoid excessive suppression of physiological ROS signaling, disruption of mitochondrial turnover, or impairment of neuronal energy metabolism. EVs intended to enhance synaptic plasticity, remyelination, cholinergic neuroprotection, or neurogenesis should be evaluated for potential effects on neural circuit stability, excitation-inhibition balance, ectopic neurogenesis, seizure threshold, and maladaptive plasticity. These mechanism-specific risks should be comprehensively assessed using behavioral testing, electrophysiology, neuropathology, mitochondrial function assays, neuroinflammatory markers, BBB integrity evaluation, and long-term cognitive monitoring.

The safety of engineered EVs should be evaluated separately from that of native EVs. Surface modification, RVG peptide display, ligand conjugation, membrane hybridization, cargo loading, and stimuli-responsive release systems may improve brain targeting and therapeutic efficacy, but may also alter EV recognition, biodistribution, clearance kinetics, immunogenicity, tissue retention, and toxicity profiles. Cargo-loaded EVs carrying miRNAs, siRNAs, mRNAs, enzymes, neurotrophic factors, mitochondrial regulators, or small molecules may induce off-target gene regulation, innate immune activation, abnormal intracellular signaling, or dose-dependent toxicity. Stimuli-responsive EV systems, such as pH-responsive, magnetic field-responsive, or light-responsive platforms, can provide spatiotemporal control over cargo release, but they may also introduce additional variables related to trigger specificity, local tissue compatibility, release repeatability, and cumulative exposure. Therefore, engineered EVs should be compared with unmodified EVs in terms of physicochemical properties, targeting efficiency, cargo-release capacity, therapeutic potency, biodistribution characteristics, immunogenicity, and repeated-dose safety.

Manufacturing quality and batch-to-batch consistency are key components of toxicological control. The properties of engineered EV preparations may vary depending on donor-cell source, oxygen tension, inflammatory priming conditions, isolation methods, purification strategies, storage conditions, and freeze-thaw cycles. Insufficiently purified or incompletely characterized EV preparations may contain protein aggregates, lipoproteins, nucleic acid contaminants, endotoxins, residual transfection reagents, culture medium components, or non-EV particles. These contaminants may lead to inconsistent efficacy or unexpected toxic reactions^[124]. Therefore, clinical-grade EV products should include preset quality control standards, including particle size distribution, particle concentration, morphology, EV-enriched marker spectrum, purity, sterility, mycoplasma-negative status, endotoxin level, residual DNA/protein content, cargo identity, potency, and stability profiles^[184]. For AD therapy, potency should be linked to disease-related functions, such as Aβ clearance, tau protein regulation, neuroinflammatory suppression, mitochondrial protection, synaptic maintenance, and neuronal survival.

Long-term and repeated-dose safety is especially important for AD because it is a chronic, progressive disease. Short-term efficacy studies may miss cumulative neurotoxicity, chronic immune dysregulation, persistent organ accumulation, delayed glial responses, or long-term effects on neural circuit function. Future research should include repeated-dose toxicity testing, dose escalation analysis, recovery assessments, long-term behavioral assessments, organ histopathology examination, CNS biodistribution studies, cognitive follow-up, and biomarker monitoring. The safety assessment of EV-based interventions in AD should go beyond the scope of general toxicity testing and adopt a structured mechanism-based framework instead. This framework should take into account factors such as donor-cell origin, cargo composition, immunogenicity, biodistribution, functional safety of the CNS, engineering-related risks, manufacturing quality, and long-term repeated dosing to define the therapeutic window of EV-based AD therapy.

Although EV-based therapy has shown important advantages for CNS drug delivery, its clinical translation in AD still faces toxicological and safety challenges. Because EVs constitute a biologically active and highly heterogeneous therapeutic category, their safety considerations extend beyond the traditional evaluation frameworks used for small-molecule drugs or conventional biologics, requiring comprehensive and mechanism-based toxicological assessment.

Beyond their therapeutic role, EVs are also emerging as promising non-invasive diagnostic tools and companion biomarkers. EVs can be isolated from a variety of biological fluids, including blood, CSF, urine, and saliva. Importantly, their lipid bilayer structure protects encapsulated proteins, nucleic acids, lipids, and metabolites from enzymatic degradation, thereby preserving disease-relevant molecular information. In the context of AD, neuron-derived or brain-enriched EVs may carry pathogenic and disease-associated molecules, including Aβ species, total tau, phosphorylated tau, inflammatory mediators, synaptic proteins, and specific miRNA signatures. This EV-associated cargo may provide a disease-relevant reflection of CNS pathological changes and may help identify patients at early or prodromal stages of AD.

From a translational perspective, EV-based biomarkers may provide important support for the clinical development and implementation of EV-based therapies. First, EV-associated Aβ, tau, and miRNA profiles may facilitate patient stratification for EV-based interventions, anti-Aβ antibody therapy, or rational combination strategies. Second, longitudinal profiling of EV cargo may allow dynamic monitoring of therapeutic responses, including amyloid clearance, attenuation of neuroinflammation, synaptic repair, and mitochondrial protection. Third, EV biomarkers can be used as an auxiliary diagnostic tool to guide the choice of personalized treatment plans, optimize the timing of treatment, and support precision medicine approaches in AD. Before EV-based diagnostic methods can be transformed into routine clinical practice, several key challenges must be addressed, including standardized EV isolation and characterization procedures to reliably enrich neuron-derived or brain-derived EV subpopulations, validated analytical platforms, and large-scale longitudinal clinical studies.

EV-based therapies for AD can be developed as independent biological intervention methods and can also be incorporated into rational combination strategies. Combining EVs with AChEIs, anti-Aβ antibodies, or lifestyle interventions may achieve broader therapeutic coverage within the complex pathological network of AD. Such combinations may simultaneously provide symptomatic relief, amyloid-targeted clearance, neuroinflammation regulation, synaptic repair, mitochondrial protection, and systemic homeostatic modulation.

EV-based therapeutic technologies have emerged as a multifunctional and biologically adaptable platform for AD therapy, with advantages in brain-targeted delivery, cargo versatility, and biocompatibility. Advances in surface engineering, cargo loading, stimuli-responsive release, and delivery-route optimization have expanded the therapeutic potential of EVs and enabled targeted modulation of Aβ pathology, tau abnormalities, mitochondrial dysfunction, and neuroinflammation in preclinical models. Early clinical studies, especially those using intranasal MSC-EVs, have provided encouraging evidence of feasibility and short-term safety. However, EV-based AD therapies still face major safety and toxicological challenges during clinical translation. Product development should prioritize comprehensive toxicological evaluation, including acute and chronic toxicity, immunogenicity, off-target accumulation, biodistribution, and risks related to repeated administration in chronic neurodegenerative diseases. The full therapeutic value of EVs will depend on the integration of advanced engineering with robust translational frameworks, including GMP-compliant manufacturing, strict quality control, standardized release criteria, and comprehensive safety profiling. Integrating these elements, together with well-designed multicenter clinical trials, will be essential for moving EV-based therapeutics from experimental research toward approved clinical use and for providing safe, effective, and scalable treatment options for patients with AD.