Engineered small extracellular vesicles as bioactive materials: Integrating engineering strategies for cargo loading and targeted delivery systems

Abstract

Small extracellular vesicles (sEVs) are increasingly regarded as a unique class of bioactive materials whose intrinsic membrane composition and nanoscale architecture provide a versatile platform for therapeutic engineering. Rather than passive carriers, sEVs can be actively programmed through diverse strategies to achieve efficient loading, precise targeting, and functional integration with synthetic systems. Endogenous modulation of donor cells—via genetic editing, priming with bioactive glass, cytokine stimulation, or hypoxic cues—enables selective packaging of nucleic acids, proteins, and metabolites into secreted vesicles. Exogenous techniques, including electroporation, sonication, and extrusion, allow controlled incorporation of therapeutic drugs or genome-editing complexes such as CRISPR/Cas. In parallel, surface modifications based on Lamp2b-fusion scaffolds, aptamers, antibodies, and click chemistry confer tissue tropism and extend circulation time. Integration with nanomaterials, scaffolds, and microfluidic platforms further enhances stability, scalability, and reproducibility, positioning sEVs at the intersection of biology and materials science. This review highlights recent advances in engineering sEVs as programmable bioactive materials and discusses their potential to transform regenerative medicine, oncology, and precision therapeutics.

Graphical abstract

Engineered sEVs as Multifunctional Bioactive MaterialsEngineering strategies transform natural sEVs into programmable nanocarriers through integrated cargo-loading and surface-modification techniques. Combining biological processes with material-based cues enhances targeting, stability, and therapeutic potency, positioning sEVs as next-generation bioactive materials for precision medicine.

Highlights

• •Engineered sEVs enable precise therapeutic delivery with enhanced biocompatibility and targeting.
• •Integrates material design, manufacturing, and regulatory roadmap for clinical translation.
• •Addresses key challenges: endosomal escape, large cargo confinement, and immune stealth.
• •Demonstrates efficacy across cancer, neurodegeneration, and regenerative models.

1. Introduction

Small extracellular vesicles (sEVs) are nano-sized extracellular vesicles (<200 nm in diameter) that mediate intercellular communication by delivering a wide array of biomolecules, including nucleic acids, proteins, lipids, and metabolites [1]. Notably, the EV field still lacks universally standardized nomenclature, as emphasized by recent guidelines [2]. In this review, we adopt the operational term “sEV” in accordance with these recommendations, using size-based classification rather than specific labels like “exosome” or “microvesicle” whose definitions often overlap. This standardized terminology is intended to improve clarity and comparability across studies. sEVs are typically released into the extracellular space through endosomal trafficking pathways, including inward budding of endosomal membranes to form multivesicular bodies (MVBs), which subsequently fuse with the plasma membrane (definition of exosomes). However, other biogenesis routes may also contribute to the sEV population, reflecting their operational classification based on size rather than definitive origin [3]. Under physiological conditions, sEVs play critical roles in maintaining homeostasis [4], while under pathological conditions, they contribute to the progression and diagnosis of diseases such as cancer, inflammation, and cardiovascular disorders [5].

The concept of extracellular vesicles has evolved over eight decades (Fig. 1). Coagulant “platelet dust” was first described in 1946; ultrastructural visualization followed in 1967. During the 1980s, investigators recognized that living cells actively secrete membrane vesicles and that such particles circulate in blood. By the late 1990s, dendritic-cell sEVs were shown to present functional MHC molecules, foreshadowing vaccine applications. The new millennium brought evidence that sEVs convey cytokines, regulate immunity, and, in a pioneering 2005 Phase-I trial, could be harnessed to elicit anti-tumor responses in patients. The founding of the International Society for Extracellular Vesicles (ISEV) in 2011 and the 2013 Nobel Prize for vesicle-trafficking research cemented the field. Recent breakthroughs—including single-vesicle imaging (2019), fast-tracked COVID-19 clinical studies (2021–2024), and the advent of engineered sEVs capable of targeted drug delivery (2025), now position sEVs at the forefront of precision nanomedicine.

Fig. 1.

Landmarks in extracellular-vesicle biology.

The concept of small extracellular vesicles (sEVs) has evolved steadily over eight decades. In 1946, Chargaff and West reported “platelet dust” with pro-coagulant activity [6]. Transmission electron microscopy first visualized these nanovesicles in 1967 [7], and biochemical characterization followed in 1974 using bat thyroid tissue [8]. By 1981 shed vesicles were recognized as discrete plasma-membrane fragments [9], and in 1983 two independent groups demonstrated that living cells actively secrete vesicles [10,11]. Circulating sEVs were identified in human blood in 1989 [12]. In 1991 vesicular release was shown to externalize obsolete membrane proteins [13]; seven years later, dendritic-cell sEVs were found to display functional MHC I/II and co-stimulatory molecules [14], and CD63⁺ sEV sub-populations were directly visualized in 1999 [15]. The first report that cytokines (for example IL-1β) can be exported via sEVs appeared in 2001 [16], followed by demonstrations that immune-cell sEVs modulate immune responses in 2003 [17,18]. A Phase-I trial in 2005 showed that tumor-peptide-loaded dendritic-cell sEVs could induce anti-tumour immunity [19]. Vesicle secretion was visualized in vivo in Caenorhabditis elegans in 2006 [20]. The International Society for Extracellular Vesicles (ISEV) was founded in 2011, and the 2013 Nobel Prize in Physiology or Medicine honored the cell-traffic machinery underlying vesicle biogenesis. Single-particle imaging began to resolve individual sEVs in live cells in 2019 [[21], [22], [23]]. Clinical translation accelerated during the COVID-19 pandemic: sEV-based trials were initiated in 2021 [24], and a double-blind randomized trial of mesenchymal stromal cell (MSC)-derived sEVs for severe COVID-19 was reported in 2024 [25]. The field now enters a “third wave”—engineered sEVs designed to surmount biological barriers and deliver precision therapeutics—heralded by landmark studies in 2025 [26].

In recent years, numerous studies have focused on exploring the drug delivery potential of sEVs, ranging from small molecules to a variety of large molecules [27]. Given their unique ability to protect cargo and mediate targeted delivery across biological barriers, sEVs have garnered significant attention as potential therapeutic delivery vehicles. Compared to synthetic nanoparticles, natural sEVs offer superior biocompatibility, reduced immunogenicity, and enhanced intracellular trafficking [28]. However, their clinical application remains limited due to inherent challenges, including rapid clearance, heterogeneous composition, and poor targeting specificity [29]. Despite their routine use, the most popular EV-isolation workflows, such as differential or density-gradient ultracentrifugation, size-exclusion chromatography, flow-field-flow fractionation, and charge- or affinity-based capture, yield vesicle preparations that vary markedly in purity, recovery and biophysical bias. The lack of harmonized, high-throughput protocols therefore remains a critical bottleneck: it undermines inter-study reproducibility and complicates GMP scale-up for therapeutic manufacturing [30].

Engineered sEVs have been developed through various strategies to enhance cargo-loading efficiency, prolong systemic circulation, and enable precise tissue targeting [[31], [32], [33]]. These approaches include manipulating donor cells for endogenous cargo enrichment [[34], [35], [36]], as well as post-isolation modification of sEV surfaces for targeted delivery [37,38]. This review aims to provide a systematic overview of current strategies for sEV engineering, with particular emphasis on loading techniques and targeting modifications. Furthermore, we highlight recent applications of engineered sEVs in treating diverse tissue-specific disorders and discuss the translational landscape, outlining key challenges and emerging opportunities in this evolving field.

2. Cargo loading

Efficient cargo loading into sEVs requires overcoming the barrier imposed by the vesicle membrane. Broadly, loading strategies can be categorized into endogenous and exogenous approaches (Fig. 2). Endogenous loading involves introducing therapeutic agents into donor cells, which are then naturally incorporated into sEVs during their biogenesis and secreted into the extracellular space. However, due to the complex intracellular environment, this method may result in non-selective cargo incorporation, including unintended co-packaging of endogenous RNAs and proteins from the parental cells [39]. Consequently, this approach may not be suitable for all cargo types. In contrast, exogenous loading involves the direct incorporation of therapeutic molecules into isolated sEVs by modulating membrane permeability. Common techniques include co-incubation, electroporation, sonication, freeze-thaw cycles, extrusion, dialysis, and chemical permeabilization using surfactants such as saponin [31]. Each technique offers distinct advantages and limitations in terms of loading efficiency, sEV integrity, and cargo stability.

Fig. 2.

Schematic overview of endogenous and exogenous cargo loading strategies for sEV engineering.

The engineering of small extracellular vesicles (sEVs) can be broadly categorized into endogenous and exogenous loading approaches. In the endogenous loading strategy (left panel), donor (parental) cells are manipulated to enrich sEV cargos during vesicle biogenesis. This can be achieved by introducing therapeutic molecules through genetic engineering (e.g., plasmid or viral vector-mediated overexpression of RNAs or proteins) or by material-based stimulation, wherein bioactive agents—such as drugs, nanoparticles, or functional biomaterials—are co-incubated with cells to modulate intracellular signaling pathways and enhance the selective packaging of desired cargos into sEVs. These cargos are sorted into vesicles via the endosomal system, including early endosomes and multivesicular bodies (MVBs), which fuse with the plasma membrane to release sEVs. This approach preserves vesicle integrity and utilizes intrinsic sorting mechanisms, but may exhibit variability in loading efficiency and specificity. In contrast, exogenous loading (right panel) involves the direct incorporation of therapeutic agents—such as molecules, RNA, or proteins—into pre-isolated sEVs by transiently disrupting their membranes using physical or chemical methods. These include electroporation, co-incubation, sonication, freeze–thaw cycles, dialysis, extrusion, and chemical permeabilization with saponin. Although exogenous loading allows precise dosing and flexible cargo choice, it may compromise vesicle stability or functional properties. Together, these strategies provide a versatile framework for engineering sEVs as targeted delivery vehicles for translational and clinical applications.

2.1. Endogenous sEV loading via donor cell modulation

A widely used endogenous cargo-loading strategy involves the modulation of donor cells, either by introducing genetic material or by stimulating them with exogenous agents, to promote the selective enrichment of therapeutic cargos within secreted sEVs. This approach takes advantage of the natural biogenesis pathway of sEVs, in which donor cells process and sort biomolecules into intraluminal vesicles within multivesicular bodies, followed by exocytosis. Cargo incorporation can occur passively or be enhanced through direct transfection, viral vectors, or lipid-based delivery systems, enabling overexpression of desired mRNAs, microRNAs, or proteins that are subsequently packaged into sEVs. For instance, several studies have successfully utilized lentiviral transduction [[40], [41], [42]], the Neon electroporation system [43], and liposome-mediated delivery [44], to enrich sEV content with specific RNAs. Extending this donor-cell modulation paradigm, overexpression of miR-17-5p achieved endogenous loading of the microRNA into sEVs that, when administered locally, restored angiogenesis and attenuated cellular senescence in diabetic wounds via PTEN/p21 pathway modulation [45].

On the other hand, another emerging strategy focuses on enriching cargo loading without directly overexpressing the desired molecules. Li et al. developed a Lamp2b-HuR and CD9-HuR (human antigen R) fusion system in donor cells, utilizing the RNA-binding domains of HuR to selectively enrich miRNA or CRISPR/dCas9 components into sEVs, thereby achieving efficient RNA inhibition both in vitro and in vivo [46,47]. Furthermore, McCann et al. revealed that intrinsic sequence motifs within miRNAs themselves can regulate their sorting into sEVs versus cellular retention [48]. Specific short sequence codes, such as EXO-motifs and CELL-motifs, were found to govern whether a given miRNA is preferentially incorporated into sEVs or retained within donor cells. This discovery offers a novel layer of control over endogenous cargo enrichment, suggesting that rational design of miRNA sequences could further enhance the selectivity and efficiency of therapeutic sEV loading. Besides, one emerging technique, cellular nanoporation (CNP), facilitates high-throughput production of mRNA-enriched sEVs by transiently permeabilizing donor cell membranes via localized electric fields. This method not only enhances mRNA uptake into donor cells but also promotes the secretion of sEVs carrying functionally active mRNA cargos [49].In a related application, CNP was used to generate let-7a-5p-loaded sEVs from Wharton's jelly-derived MSCs, which significantly alleviated pulmonary inflammation and fibrosis in an acute lung injury model [50].

Beyond nucleic acid-based manipulation, exogenous stimuli have been shown to modulate donor cell behavior and sEV content. Co-incubation of stem cells with bioactive glass [51,52], inflammatory cytokines [53], hypoxia [54], and H₂O₂ [55] enhances the production of sEVs with regenerative properties, highlighting the role of microenvironmental cues in guiding sEV biogenesis and cargo sorting. Xu et al. demonstrated that priming donor cells with bioactive-glass (BG) extract significantly alters their secretome, yielding BG-conditioned sEVs with enhanced abilities in macrophage polarization and angiogenesis [51]. Consistent with this observation, Wu et al. reported that direct BG incubation boosts the number and pro-angiogenic potency of stem-cell-derived sEVs [52]. Beyond biomaterial cues, Lee et al. demonstrated that the regenerative capacity of MSCs can be mediated via an autocrine signaling loop by their secreted sEVs, and the therapeutic efficacy can be licensed by different inflammatory cytokines [53]. Likewise, hypoxic pre-conditioning of MSCs increases the yield of cardio-protective sEVs enriched in miR-125b, which limits apoptosis and promotes repair after myocardial infarction [54].

This donor cell modulation strategy offers several notable advantages. It preserves the structural integrity and membrane protein composition of sEVs, which is essential for maintaining their natural targeting properties and biocompatibility. Moreover, it enables cell-intrinsic cargo selection, allowing for the endogenous packaging of labile or sensitive molecules such as RNAs and enzymes. The approach is also biologically relevant and scalable, as it mimics physiological or pathological stimuli that naturally regulate sEV biogenesis and cargo sorting. Nevertheless, donor cell modulation presents important limitations. Cargo loading efficiency often remains suboptimal, and enrichment specificity is limited by the complexity and heterogeneity of intracellular sorting pathways. Additionally, prolonged stimulation or genetic modification of donor cells may lead to phenotypic drift or genomic instability, posing challenges for reproducibility and clinical translation. Batch-to-batch variability due to differences in transfection efficiency, cell source, and culture conditions further complicates standardization. Despite these challenges, this approach remains a cornerstone of sEV engineering and continues to evolve with the integration of high-throughput screening, omics-guided profiling, and synthetic biology techniques to enhance both precision and translational feasibility.

2.2. Co-incubation-mediated cargo loading into sEVs

In the exogenous co-incubation approach, purified sEVs are directly mixed with therapeutic agents. This strategy can be primarily divided into two subcategories: (i) Passive Diffusion: sEVs are incubated with cargo under conditions that promote spontaneous diffusion across the membrane, driven by concentration gradients. The mechanism is primarily attributed to the hydrophobic and lipid-rich nature of the sEV bilayer, which enables spontaneous incorporation of lipophilic compounds. Meanwhile, the aqueous core of sEVs allows for the encapsulation of hydrophilic cargos, expanding the range of molecules that can be accommodated [[56], [57], [58]]. (ii) Assisted Loading: For impermeable macromolecules such as nucleic acids, the loading efficiency can be significantly enhanced by employing membrane permeability enhancers. This approach is particularly suitable for nucleic acid loading, as several studies have reported the successful incorporation of siRNA and miRNA into sEVs using commercial transfection reagents [59,60]. These methods are typically mild and do not compromise the structural integrity of sEV membranes, making them favorable for preserving vesicle functionality during the loading process.

In addition to nucleic acids, small molecules such as doxorubicin have also been passively incorporated into sEVs from different cell types derived from MSCs and macrophages using this method [61,62]. While this approach is technically straightforward and preserves sEV integrity, the loading efficiency is relatively unsatisfactory and highly dependent on the physicochemical properties of both the sEVs and the cargo. Haney et al. found that pH modulation significantly influenced the diffusion of hydrophilic compounds across the sEV membrane, which can facilitate their incorporation [63]. In general, hydrophilic cargos tend to accumulate in the aqueous lumen, whereas hydrophobic cargos preferentially integrate into the lipid bilayer [56,64].

Beyond conventional small molecules, engineered nanomaterials can also be introduced into sEVs despite their lack of an active endocytic system like cells. Loading of such materials relies primarily on surface conjugation mechanisms or passive physicochemical interactions, including electrostatic, hydrophobic, or receptor-mediated processes. For instance, Qi et al. reported that transferrin-functionalized Fe₃O₄ nanoparticles could be anchored onto the sEV membrane through receptor–ligand interactions involving transferrin receptors [65]. In another study, Betzer et al. demonstrated that glucose-coated gold nanoparticles were effectively internalized by MSC-derived sEVs via GLUT1-mediated uptake, implicating membrane-associated transporter mechanisms [66].

These findings suggest that cargo composition, surface chemistry, and environmental parameters—such as pH, ionic strength, and membrane fluidity—play critical roles in determining loading efficiency during co-incubation. Accordingly, systematic optimization of these factors is essential to maximize yield and maintain sEV functionality for translational applications.

2.3. Electroporation-mediated cargo loading into sEVs

Electroporation is a widely utilized technique for exogenous cargo loading into sEVs, involving the application of short-duration, high-voltage electrical pulses that induce transient pores in the vesicle membrane. These temporary disruptions in the lipid bilayer enable hydrophilic molecules, including nucleic acids and small-molecule drugs, to enter the sEV lumen. Upon cessation of the electric field, the membrane reseals, thereby encapsulating the cargo within the sEVs.

Several studies have demonstrated the versatility of electroporation for diverse cargo types and therapeutic applications. For example, co-electroporation of chemotherapy drug (5-fluorouracil) and miRNA inhibitor into sEVs results in enhanced apoptosis in colon cancer cells [67]; electroporation of siRNA targeting oncogene Src into sEVs induces apoptosis of both senescent stromal cells and tumor cells [68]. Gu et al. incorporated miR-302 mimics into engineered sEVs and demonstrated improvements in cardiac function, reduced infarct size, and attenuation of myocardial apoptosis and inflammation in a myocardial infarction model [69]. Beyond nucleic acids, electroporation has also been applied to load immunogenic proteins and small molecules. For instance, Huang et al. loaded human neutrophil elastase (ELANE) and the TLR3 agonist Hiltonol into α-lactalbumin-engineered breast cancer-derived sEVs to generate HELA-sEVs, which served as an in situ dendritic cell vaccine. These HELA-sEVs exhibited robust antitumor activity in murine models and human breast cancer organoids by enhancing cDC1 activation and promoting tumor-reactive CD8⁺ T cell responses [70]. Xu et al. demonstrated that loading Kartogenin (KGN), a TGF-beta/Smad activator, into sEVs by electroporation enhances chondrogenesis of MSCs for cartilage regeneration [71]. Besides, Wan et al. established a versatile CRISPR delivery platform, "exosomeRNP," by loading pre-assembled Cas9-sgRNA ribonucleoproteins (RNPs) into LX-2 human hepatic stellate cell (HSC)-derived sEVs. Capitalizing on the innate liver tropism of these vesicles, intravenously injected exosomeRNPs efficiently accumulated in the liver and executed therapeutic genome editing. The platform's broad utility was validated across multiple disease models, including ameliorating acute liver injury by targeting PUMA, attenuating chronic fibrosis via CCNE1 modulation, and suppressing hepatocellular carcinoma through KAT5 editing, collectively confirming its safety and efficacy as a liver-directed CRISPR strategy [72] (Fig. 3). Recently, novel electroporation-based engineering strategies have been developed to enhance the therapeutic functionality and targeting specificity of sEVs. For instance, Dong et al. introduced an adaptive electroporation-integrated microfluidic platform to generate CD64-expressing sEVs, which enable high-affinity binding with therapeutic antibodies. These antibody-decorated sEVs were employed to deliver IFN-γ mRNA to glioblastoma cells, eliciting potent anti-tumor effects in both drug-resistant tumor models and patient-derived organoids [73]. In a separate approach, a newly developed electroporation-guided cargo removal technique was applied to eliminate oncogenic contents from cancer-derived sEVs while preserving their intrinsic homing ability. The resulting vesicles were fused with liposomes via a membrane fusion process to form engineered hybrid nanoparticles, which were subsequently modified with anti-EGFR monoclonal antibodies to enhance cancer cell targeting. These multifunctional hybrids exhibited significantly improved drug loading capacity and therapeutic efficacy, while simultaneously addressing the biosafety concerns associated with cancer cell-derived sEVs [74].

Fig. 3.

(A) Schematic of exosome-based delivery of Cas9 RNP for treating liver diseases. LX-2 (human hepatic stellate cell, HSC)-derived sEVs are purified, electroporated with pre-assembled Cas9–sgRNA RNPs, and delivered by IV injection. Leveraging the intrinsic liver tropism of HSC sEVs, the nanocomplexes accumulate in the liver and execute therapeutic genome editing in vivo. Disease-specific interventions: PUMA targeting ameliorated acute liver injury; CCNE1 (Cyclin E1) modulation attenuated chronic liver fibrosis; and KAT5 editing suppressed orthotopic hepatocellular carcinoma-collectively validating exosomeRNP as a safe, liver-addressable CRISPR delivery strategy. (B) Exosome-mediated Cas9 RNP delivery enabling genome editing. (C–F) Characterization of purified exosomes. (G–I) Analysis of exosome–RNP complexes. (J & K) Exosome-driven cytosolic delivery of Cas9-FITC in LX-2 and Huh-7 cells. Copyright 2022, The American Association for the Advancement of Science.

Notably, commonly used donor cell types include HEK293T cells, MSCs, and dendritic cells. However, the applied electroporation voltages vary widely, and loading efficiencies remain inconsistent, ranging from as low as 0.5 % [67] to as high as 83 % [75]. These discrepancies highlight the need for standardized electroporation protocols to optimize cargo delivery outcomes while maintaining sEV integrity and function. Table 1 provides detailed information on recent publications that employ electroporation to load diverse cargos into sEVs, supplementing the representative studies discussed above.

Table 1.

Electroporation-mediated loading strategies for engineered sEVs.

Therapeutic cargo	Cell source of sEVs	Voltage	Load efficiency	Target diseases	Function	Ref.
FAK siRNA	HEK293T cells	800V	/	Colon cancer	Activate paraptosis and inhibit the proliferation of drug-resistant colon cancer cells. Suppressed tumor growth and metastasis.	[76]
IL-12 mRNA	HEK293T cells	200V	27.6 %	Lung carcinoma	An inhaled sEVs platform delivers IL-12 mRNA selectively to lung cells, enabling localized protein expression that suppresses tumors while minimizing systemic exposure.	[77]
AgomiR-23b	Bone marrow MSCs	200V	32.2 %	Sepsis-related acute lung injury	Intratracheal sEV-encapsulated miR-23b mimics alleviate ALI by suppressing M1 macrophage activation via the Lpar1-NF-κB pathway.	[78]
HMOX1 siRNA	Cardiomyocyte HL-1	700V	/	DOX-induced cardiomyopathy	Block the ferroptosis and the subsequent cardiotoxicity induced by doxorubicin.	[79]
VEGFA-GFP mRNA	Neonatal human dermal fibroblasts	200V	/	Bone defects	A sEV-GelMA hybrid hydrogel enables spatiotemporally controlled delivery of VEGFA mRNA, synergistically enhancing angiogenesis and osteogenesis while upregulating key osteogenic markers (ALP/RUNX2/OPN) in pre-osteoblasts.	[80]
CRISPR-Cas9 RNP	Hepatic stellate cells	/	20 %	Liver diseases	Facilitate effective cytosolic delivery of RNP and treat acute liver injury, chronic liver fibrosis, and hepatocellular carcinoma.	[72]
Drp1 siRNA	HEK293T cells	700V		Pulmonary fibrosis	Engineered EVs surface-functionalized with D-mannose and loaded with siDrp1 attenuate pulmonary fibrosis by targeting Mømitohigh macrophages-following ECM-priming by MMP19-exosomes-to inhibit pathological mitochondrial fission.	[81]
Mitochondria-targeting moiety TPP-PDL	Expi293F cells	700V	/	Ischemic stroke	The dual-targeting RVG-EVmt system delivers circMTCO2 to neuronal mitochondria, suppressing mtROS-induced ferroptosis to reduce cerebral infarct volume and restore neurological function.	[82]
FRATtide	Escherichia coli Nissle 1917	100V	21.5 %	Osteoporosis	Engineered bacterial vesicles (BEV-DCS) from DC-STAMP-expressing E. coli enable targeted FRAT delivery to pre-osteoclasts, which demonstrated bone-specific targeting, ameliorated osteoporosis, and exhibited no toxicity.	[83]
miR-21 inhibitor (miR-21i) 5-Fluorouracil (5-FU)	HEK293T cells	1000V	miR-21i: 0.5 %, 5-FU: 3.1 %	Colorectal carcinoma	Reverse drug resistance and enhance cytotoxicity in colon-cancer cells	[67]
miR-54	HEK293T cells	700V	/	Obesity	Induce the local omental adipose tissue browning	[84]
Src siRNA	iMSCs (derivates of induced pluripotent stem cells	400 mV	45 %	Cancer	Anti-cancer, eliminate senescent stromal cells and tumor cells	[68]
miR-302 mimic	Bone MSCs	350V	/	Myocardial infarction	Improve cardiac function, attenuate myocardial apoptosis and inflammatory response, and reduce infarct size.	[69]
Catalase mRNA	HEK293T cells	1200 mV	/	Parkinson's disease	Rescue neurotoxicity and attenuate of ROS-triggered neuroinflammation	[85]
Triptoline	Human Rheumatoid arthritis (RA) fibroblasts	400V	81.22 %	Rheumatoid arthritis	Inhibit the proliferation of RA fibroblasts	[86]
siGPX4	Human MSCs	/	/	Glioblastoma	Disrupt DHODH, disable GPX4 defense, and trigger Fe2+ release via MNPs	[87]
Hydrochloride	M2 macrophages	100V	25.14 %	Atherosclerosis	Generate anti-inflammatory carbon monoxide and bilirubin to enhance the anti-inflammatory effects	[88]
tLyp-1 siRNA	HEK293T cells	400V	61.53 %	Lung cancer	Reduce the stemness of cancer stem cells.	[89]
Nefmut-Tat protein	Dendritic cells	1300V	32.8 %	HIV	Elicit strong Th1/CTL responses and sustain protection against SCR HIV-1 exposure	[90]
Human neutrophil elastase (ELANE) and Hiltonol (TLR3 agonist)	HEK293T cells	400V	ELANE: 12.45 %, Hiltonol: 21.31 %	Breast cancer	Induce tumor-specific immunogenic cell death in breast-cancer cells	[70]
VEGF plasmid DNA	ATDC5, a mouse chondrogenic progenitor cell line	300V	83 %	Cranial defect	Sustainably deliver VEGF gene and enhance vascularized osteogenesis	[75]
Imiquimod, an immune adjuvant	HEK 293T cells	400V	12 %	Melanoma; Breast cancer-	Promote dendritic cell maturation and restore CD8+ T-cell function	[91]
siShn3	iMSCs (derivates of induced pluripotent stem cells	700V	18 %	Osteoporosis	Enhance osteogenic differentiation, suppress RANKL-driven osteoclastogenesis, and up-regulate SLIT3 to stimulate type-H vascularization	[92]
siMMP13	Expi293F cells	450V	34.96 %	Osteoarthritis	Lower MMP-13, raise COL2A1/proteoglycan, and normalize IL-1β-perturbed chondrocyte proteins	[93]
Oxaliplatin	HEK293T cells	1000V	/	Colon cancer	Reverse drug resistance and suppress tumor growth	[94]
Lamp2b/HuR fusion protein	HEK293T cells	400V	/	Liver fibrosis	Alleviate liver fibrosis by down-regulating miR-155 and other inflammatory genes	[46]
miR-26a	HEK293T cells	400V	/	Liver cancer	Up-regulate miR-26a to curb cell migration, proliferation and cell cycle progression	[95]
miR-140	Dendritic cells	350V	60 %	Osteoarthritis	Suppress cartilage-degrading proteases to slow osteoarthritis progression	[96]
Cas9 protein/sgRNA	HEK293T cells	/	36.99 %	Glioblastoma	Achieve high-precision glioma targeting, deep tumour penetration and efficient, off-target-free gene editing	[97]
ANCR plasmid	Human bone marrow MSCs	100V	/	Calcification of autologous pathological vessels	Accelerate smooth-muscle reconstruction while blocking Gli1+ transdifferentiation and kidney-disease-induced vascular calcification	[98]
SIRT6 siRNA	HEK293T cells	400 mV	/	Prostate cancer	Inhibit tumor growth and metastasis.	[99]
Doxorubicin	Dendritic cells	350V	20 %	Breast cancer	Halt tumor expansion without systemic toxicity	[100]
Kartogenin	Dendritic cells	/	40 %	Osteoarthritis	Drive chondrogenesis.	[71]
Liposomes	MCF-7 cell line	350V	52.23 %	Breast Cancer	Deliver anticancer drugs at high payload, with superior targeting, therapeutic efficacy and minimal cytotoxicity	[74]

∗Additional recent publications are included in the table, thereby complementing the representative examples discussed in the main text.

2.4. Sonication-mediated cargo loading into sEVs

Sonication is another utilized physical loading technique that enhances membrane permeability through the application of mechanical shear force. The transient disruption of the sEV lipid bilayer facilitates the incorporation of diverse therapeutic cargos, including small-molecule drugs, proteins, nucleic acids, and nanomaterials [101]. A growing number of studies have adopted sonication to achieve high-efficiency cargo loading while preserving vesicle functionality (Table 2).

Table 2.

Sonication of engineered sEVs as cargo loading strategies.

Therapeutic cargo	Cell source of sEVs	Load efficiency	Target diseases	Function	Ref.
Erastin;	HEK293T cells	Er: 84 %,	Cancer	Induce ferroptosis under 532 nm laser irradiation while minimizing liver toxicity	[102]
Rose Bengal		RB: 60 %
Doxorubicin	B16F10 cells	/	Melanoma	Suppress melanoma growth	[103]
Corynexin-B	HT22 cells	45 %	Alzheimer's disease	Trigger autophagy in APP-expressing neurons, mitigating Alzheimer-like pathology and cognitive decline	[104]
Quercetin	Whole blood	13.6 %	Cerebral ischemia/reperfusion injury	Protect neurons by activating the Nrf2/HO-1 axis and reducing ROS	[105]
Minoxidil	Dermal papilla cells	20.24 %	Androgenetic alopecia	Enlarge and increase dermal hair follicles, stimulate perifollicular angiogenesis, and drive telogen-to-anagen transition	[106]
TGF-β1 and IL-10	Human monocyte-derived dendritic cells	/	Periodontitis	Up-regulate PD-L1 while down-regulating CD86, HLA-DR and CD80	[107]
Prussian blue nanoparticles	Neutrophils	/	Rheumatoid arthritis	Home to activated synoviocytes, neutralize pro-inflammatory mediators, scavenge ROS and relieve joint inflammation	[108]
Curcumin	Bone marrow MSCs	18.9 %	Smoking-related osteoporosis	Restore the osteogenic differentiation potential of BMSCs and mitigating bone loss	[110]
Lonicera japonica	Bone marrow MSCs	71.2 %	Cancer	LJP-EVs-primed dendritic cells potentiate tumor-reactive CD8+ T cell responses, effectively suppressing tumor growth in immunologically cold models.	[111]
Thunb. Polysaccharides (LJP)
Polydeoxyribonucleotide (PDRN)	Tea	35 %	Inflammatory bowel disease	Tea-derived EVs act as oral nanocarriers for PDRN delivery, which demonstrate gastrointestinal stability and achieve targeted anti-inflammatory effects in colitis through macrophage M2 polarization via the cAMP/HIF-1α pathway.	[112]

∗Additional recent publications are included in the table, thereby complementing the representative examples discussed in the main text.

Several studies have demonstrated the efficacy of sonication in facilitating the encapsulation of small-molecule therapeutics into sEVs. Du et al. sonicated HEK293T-derived sEVs in the presence of the ferroptosis inducer Erastin (Er) and the photosensitizer Rose Bengal (RB), achieving remarkable loading efficiencies of 84 % and 60 %, respectively [102]. Kang et al. applied sonication to generate doxorubicin-loaded sEVs, which displayed superior antitumor efficacy against melanoma both in vitro and in vivo [103]. In a regeneration context, Corynexin-B was incorporated into HT22 cell-derived sEVs with a 45 % loading efficiency, following a controlled sonication protocol (20 % amplitude, 6 cycles of 30s on/150s off) [104]. Guo et al. loaded Quercetin into sEVs via sonication for neuronal delivery, resulting in reduced oxidative stress and enhanced neuronal survival through Nrf2/HO-1 pathway activation [105]. Additionally, Zhang et al. utilized sonication to load Minoxidil into dermal papilla cell-derived sEVs, promoting hair follicle enlargement, capillary formation, and induction of anagen-phase transition in hair follicle cycling [106]. Sonication has also been applied for the incorporation of biologically active proteins into sEVs. Human monocyte-derived dendritic cell (moDC)-sEVs were loaded with TGF-β1 and IL-10 via sonication, resulting in sEVs with robust anti-inflammatory activity in a preclinical model of periodontitis [107]. These protein-loaded sEVs significantly attenuated local inflammation, suggesting a viable strategy for immune modulation. Additionally, sonication provides a physical means to incorporate inorganic and organic nanomaterials into sEVs for diagnostic or theranostic applications. Zhang et al. successfully embedded Prussian blue nanoparticles (PBNPs) into sEVs using sonication, enabling antioxidant and photothermal therapeutic effects [108]. On the other hand, Nucleic acid cargos can also be efficiently loaded into sEVs using ultrasonic energy. Hu et al. reported the incorporation of miR-378 into sEVs via a controlled ultrasonic shock method, demonstrating successful delivery and functional activity of the encapsulated miRNA in downstream assays [109].

Despite its broad applicability, sonication remains limited by the lack of standardized parameters, with few studies reporting critical details such as amplitude, cycle duration, or temperature control factors essential for reproducibility and sEV integrity [104]. While sonication offers a robust, accessible, and efficient means of loading hydrophobic or membrane-active cargos with relatively high efficiency, its non-specific mechanical force may disrupt vesicle morphology or compromise bioactivity if not optimized. The absence of consensus protocols also hinders scalability and regulatory translation. Future efforts should aim to establish reproducible sonication conditions, potentially integrating real-time membrane monitoring or microfluidic modulation. When properly controlled, sonication remains a valuable strategy for generating clinically relevant, cargo-loaded sEVs.

2.5. Freeze-thaw cycle-mediated cargo loading into sEVs

The freeze-thaw method, originally developed for lysing bacterial and mammalian cells or facilitating intracellular delivery, has been repurposed in the context of extracellular vesicle engineering. In this approach, sEVs are mixed with therapeutic cargos and subjected to repeated cycles of rapid freezing (typically at −80 °C) and thawing at room temperature. The formation of ice crystals during freezing transiently disrupts the phospholipid bilayer, allowing passive diffusion of cargo into the vesicular lumen. It is generally recommended that at least three freeze-thaw cycles be performed to enhance membrane permeabilization.

However, as summarized in Table 3, standalone freeze-thaw strategies are associated with relatively low encapsulation efficiencies [113], likely due to limited membrane destabilization and potential cargo degradation. To address these limitations, Ebrahimian et al. employed a hybrid approach combining co-incubation, surfactant treatment, and freeze-thaw cycling to encapsulate Thymoquinone into sEVs, achieving a 60 % loading efficiency and improved anti-cancer effects in breast cancer models [114]. Similarly, Tran et al. integrated incubation, sonication, and freeze-thaw cycles to achieve up to 80 % loading efficiency, demonstrating the advantage of multimodal enhancement strategies [115]. Singh et al. reported a membrane fusion strategy that merges BMSC exosomes with PPyNP-loaded liposomes, creating a unified vesicle which combines native biological payloads with introduced conductive properties. This hybrid system introduces a tunable bioelectric capability for electro-assisted regeneration while better preserving native membrane proteins than chemical methods. Furthermore, the one-pot, solvent-free process is scalable and provides built-in photophysical readouts for quality control, establishing a modular platform adaptable to diverse functional nanoparticles [116] (Fig. 4).

Table 3.

Freeze-thaw cycles of engineered sEVs as cargo loading strategies.

Therapeutic cargo	Cell source of sEVs	Cycles	Load efficiency	Target diseases	Function	Ref.
Polypyrrole nanoparticles	Bone marrow MSCs	Liquid nitrogen ∼50 °C, 15 min/cycle,10 times	/	Diabetic neuropathies	Restore gastrocnemius morphology and mass, control hyperglycemia and body-weight loss, and lessen pancreatic, renal and hepatic injury	[116]
CD47-overexpressed hybrid therapeutic nano- vesicles	CT26 cells	−80 °C–37 °C, 3 times	/	Cancer	Block CD47 to boost macrophage phagocytosis, achieve laser-activated photothermal ablation, and provoke immunogenic tumor cell death with robust antigen release	[117]
Thymoquinone	Human adipocyte-derived MSCs	−80 °C–37 °C, 3 times	60 %	Breast cancer	Selectively lower MCF-7 breast cancer viability while sparing normal L929 cells	[114]
Catalase	RAW264.7 cells	−80 °C ∼ Room temperature, 3 times	14.7 %	Parkinson's disease	Provide neuroprotection via catalase-loaded sEVs that show high loading efficiency, protease resistance and sustained enzyme release	[113]
Aspirin (POX 407 and TPGS)	HT29 cells	−20 °C 18 h followed by Room temperature 1 h	80 %	Breast and colorectal cancer	Increase aspirin encapsulation, solubility and cytotoxic potency in cancer cells	[115]
Aspirin (POX 407 and TPGS)	HT29 cells	−20 °C 18 h followed by Room temperature 1 h	/	Breast and colorectal cancer	Enhance aspirin uptake through clathrin-dependent and independent routes, thereby intensifying apoptosis and autophagy in breast and colorectal tumors	[118]

∗Additional recent publications are included in the table, thereby complementing the representative examples discussed in the main text.

Fig. 4.

(A) Schematic of exosome–liposome membrane fusion via freeze-thaw. Engineered fused exosomes via freeze–thaw membrane fusion of BMSC-derived exosomes with anionic liposomes pre-encapsulating polypyrrole nanoparticles (PPyNPs). Fusion reorganizes both bilayers into a single vesicle that co-entrapps PPyNPs, coupling native miRNA/protein payloads with a conductive biophysical cue; R18 dequenching ≈28 %. This strategy adds a tunable bioelectric modality for electro-assisted regeneration, preserves membrane proteins/tropism better than harsh chemistries, and offers modular conductivity control by PPy dose. It also provides built-in FTIR/UV–Vis/electrical readouts useful as PAT/lot-release surrogates and is a one-pot, solvent-free, scalable process extensible to other functional nanoparticles (photothermal, magnetothermal, imaging). (B1) TEM images of fused structures. Yellow: exosome; Black: liposome; Red: anti-CD9-gold. (B2) NTA of exosome–polypyrrole liposome fusions. (C) Venn diagram of LC-MS peaks: freeze-thawed vs. untreated exosomes. (D1) PKH-26-labeled exosomes. (D2) PKH-67-labeled polypyrrole liposomes. (D3) Fused exosome–liposome complexes. (D4) FT-IR spectra of polypyrrole nanoparticles. (D5) UV–vis absorption of polypyrrole at varying concentrations. Copyright 2021, Elsevier.

Despite its technical simplicity and accessibility, the freeze-thaw method presents several limitations, including low cargo retention, vesicle aggregation, and potential protein denaturation due to thermal stress. Future refinements may involve the use of cryoprotectants or kinetic modulation of the thawing process to preserve vesicle integrity while maximizing cargo incorporation.

2.6. Dialysis-mediated cargo loading into sEVs

Dialysis-based cargo loading represents a gentle, membrane-driven method that enhances the permeability of sEV membranes through osmotic modulation. In this approach, sEVs are mixed with therapeutic cargos in a hypotonic buffer solution and placed within a dialysis membrane. The hypotonic environment induces osmotic swelling of sEVs, transiently stretching the lipid bilayer and creating pores that facilitate passive diffusion of cargo molecules into the vesicle lumen. Additionally, the concentration gradient across the dialysis membrane serves as a driving force to enhance cargo uptake.

Wei et al. demonstrated that doxorubicin-loaded sEVs prepared via dialysis were effectively internalized by cancer cells and significantly inhibited tumor cell proliferation in vitro [119]. In a comparative study, Fuhrmann et al. reported that dialysis-mediated loading achieved more than an 11-fold increase in encapsulation efficiency compared to passive incubation, underscoring its potential for enhanced drug delivery applications [64]. Jeyaram et al. developed a force-free method for efficient nucleic acid encapsulation into EVs by pre-establishing a stable pH gradient. The process involves programming the EVs with an acidic intraluminal pH (via dehydration/rehydration in citrate buffer, pH 2.5) and a neutral exterior, followed by cargo incubation which drives protonation and intraluminal trapping. Optimized at ∼22 °C for 2 h, this tunable and scalable strategy avoids vesicle damage, minimizes cargo aggregation, and maintains EV integrity and bioactivity, offering a robust alternative to harsh physical loading methods [120] (Fig. 5).

Fig. 5.

(A) Schematic of pH-EVs preparation and loading via dialysis. A force-free, pH-gradient–driven approach enables efficient nucleic-acid encapsulation into EVs. EVs are first dehydrated in 70 % ethanol (∼12 h), then rehydrated in 150 mM citrate buffer at pH 2.5 (∼1 h) and dialyzed against HEPES-buffered saline at pH 7 (∼24 h) to establish an acidic intraluminal pH while maintaining a neutral exterior. Subsequent incubation of cargo with EVs (e.g., 3 × 10⁹ EVs with ∼1000 pmol nucleic acid) drives protonation/ion-pairing and intraluminal trapping across the pH gradient. Optimization identified ∼22 °C and ∼2 h as favorable loading conditions, with internal pH ∼2.5 yielding maximal association; prolonged incubation (≥6 h) reduced loading. This pH-programming method avoids sonication/electroporation, reduces vesicle damage and cargo aggregation, provides tunable and reproducible parameters (temperature, time, lumenal pH) amenable to process-analytical control, and offers a scalable route to functional RNA loading while retaining EV integrity and bioactivity. (B) EV concentration measurement via NTA. (C) Protein-associated EV quantification with/without pH modification. (D&E) Protein and immunoblot analysis of unmodified/modified EVs. (F) Quantitative immunoblotting of EV markers. (G) TEM imaging of EV morphology. (H–J) miR-93 loading efficiency under varying conditions. Copyright 2020, Elsevier.

However, despite its technical advantages, the dialysis method also presents several limitations. Extended exposure to hypotonic stress may lead to protein degradation, reducing the stability or activity of sensitive therapeutic cargos [120]. Furthermore, from a translational perspective, the complete removal of hypotonic buffer components and dialysis residues is essential to ensure the safety and compatibility of the final sEV formulation, particularly for clinical applications. Variations in dialysis buffer composition and molecular weight cut-offs also introduce potential batch-to-batch variability, which may complicate regulatory compliance and manufacturing standardization. Representative studies employing dialysis-mediated sEV loading are summarized in Table 4.

Table 4.

Dialysis of engineered sEVs as cargo loading strategies.

Therapeutic cargo	Cell source of sEVs	Load efficiency	Target diseases	Function	Ref.
Doxorubicin	Bone marrow MSCs	12 %	Osteosarcoma	Boost cellular uptake and anti-tumour potency in MG-63 osteosarcoma cells while sparing H9C2 cardiomyocytes	[119]
Nucleic acid	HEK293T cells	6.5 %	/	Maintain normal EV uptake with no detectable cytotoxicity	[120]
Doxorubicin and immunoglobulin G	Hela, MDA-MB-231, and human brain microvascular endothelial cells	85 %	Neurological diseases	A membrane fusion-based method that utilizes fusogenic lipid nanoparticles (cubosomes) to non-destructively load large molecules into EVs.	[121]
Bortezomib (Btz)	Monocytes	13.2 %	Multiple myeloma	Monocyte-derived sEVs achieved targeted bortezomib delivery into multiple myeloma cells, enabling sustained drug accumulation and enhanced intracellular accessibility.	[122]
Porphyrins	MDA-MB231 breast cancer (MDA) cells, human umbilical vein endothelial cells, human bone-marrow derived MSCs, human embryonic stem cells	>60 %	Cancer	Amplify the photodynamic and phototoxic effects of hydrophobic porphyrins in cancer models	[64]

∗Additional recent publications are included in the table, thereby complementing the representative examples discussed in the main text.

2.7. Extrusion-mediated cargo loading into sEVs

Extrusion is a mechanical approach that enhances sEV cargo loading by forcing a mixture of sEVs and therapeutic agents through nanoporous membranes, typically 100–400 nm in diameter. This process creates transient disruptions in the membrane, allowing drug molecules to diffuse into or associate with the vesicles. Cargo is typically dissolved in a suitable solvent and blended with sEVs prior to extrusion using a manual extruder or microgrinder.

This technique has been used to load small-molecule drugs such as doxorubicin into sEVs or sEV-mimetic vesicles, achieving loading efficiencies ranging from 8.5 % to 68 % and demonstrating enhanced anti-tumor activity in preclinical models [123,124]. Guo et al. developed an endosome-targeted magnetic extrusion workflow for the scalable production of endosome-derived vesicles (EMs). This process, which involves magnetically enriching and isolating endosomes before extrusion, achieves high doxorubicin loading efficiency (∼68 %) under an ammonium sulfate gradient—significantly outperforming passive loading. The method provides orders-of-magnitude higher particle yields than conventional methods while preserving EV-like properties, offering a standardized and scalable solution to clinical manufacturing challenges [125] (Fig. 6). In addition to improved drug delivery, extrusion offers benefits in scalability and vesicle size uniformity. However, extrusion may alter vesicle size and compromise the structural integrity and native protein composition of sEVs, as it often generates vesicle mimetics rather than naturally secreted sEVs. These alterations could affect vesicle biodistribution, cellular uptake, and immunogenicity in vivo, thereby influencing therapeutic efficacy and safety. Relevant studies using extrusion-mediated cargo loading are summarized in Table 5.

Fig. 6.

(A) Schematic of the magnetic extrusion procedure from endocytosis to magnetic purification of exosome mimetics (EMs). An endosome-targeted magnetic extrusion workflow produces EMs at scale: cells first endocytose ∼10 nm iron-oxide nanoparticles (IONPs) to magnetically enrich endosomes; organelles are released and magnetically isolated, then extruded through ∼200 nm nanoporous membranes to yield EV-sized vesicles. Under a 240 mM ammonium-sulfate transmembrane gradient during extrusion, doxorubicin achieved a loading efficiency of ∼68 % at 1 mg/mL input, whereas passive loading yielded only ∼20–25 %. Loading efficiency increased monotonically as the input concentration rose from 0.2 to 1 mg/mL. The method delivers order-of-magnitude higher particle yields than conventional EV isolation while preserving EV-like size/markers and in vivo drug-carrier performance, providing a potential standardizable, scalable route that directly addresses CMC bottlenecks for clinical translation. (B) TEM images of IONP-loaded cells, endosomes, and EMs; NTA of EM and EV samples. (C) NTA profiles of 231-EM, 231-EV, 3T3-EM, and 3T3-EV. (D&E) Yield and protein content of EMs versus EVs from two cell lines. (F&G) Western blot of marker proteins in EMs, EVs, and cell lysates. Copyright 2021, John Wiley and Sons.

Table 5.

Extrusion of engineered sEVs as cargo loading strategies.

Therapeutic cargo	Cell source of sEVs	Load efficiency	Target diseases	Function	Ref.
Doxorubicin	Neutrophils	8.5 %	Cancer	Suppress tumor-cell proliferation more effectively	[123]
Dihydrotanshinone I-loaded nanocarrier	MSCs	39.67 %	Parkinson's disease	Home to CCL2-rich substantia nigra, block peripheral immune infiltration and activate the Nrf2–GPX4 axis to curb microglial ferroptosis and inflammation	[126]
A near-infrared II (NIR-II) organic molecule (MYM)	Expi293F cells	49.5 %	Glioblastoma	An iRGD-engineered sEVs platform encapsulating the NIR-II molecule MYM, which enables precise glioblastoma targeting through blood–brain barrier penetration and elicits potent photothermal and immunotherapeutic responses under laser activation.	[127]
Doxorubicin	U937 monocytic cells	/	Cancer	Induce TNF-R-dependent endothelial death in a dose-responsive manner; drug-loaded nanovesicles localize to tumors and retard growth with fewer systemic side-effects than free drug	[124]
sEVs mimetics	Natural killer cells	/	Cancer	Reduce tumor bioluminescence, volume and mass	[128]
Doxorubicin	Fibroblast 3T3 cells	68 %	Breast cancer	Deliver doxorubicin to orthotopic breast tumors, markedly inhibiting their growth	[125]

∗Additional recent publications are included in the table, thereby complementing the representative examples discussed in the main text.

2.8. Saponin-mediated cargo loading into sEVs

Saponin, a commonly used surfactant, facilitates cargo loading by interacting with membrane cholesterol and inducing pore formation in the sEV lipid bilayer, thereby increasing membrane permeability [129]. It is often employed in combination with other loading strategies, such as microfluidic platforms or freeze–thaw cycles, to further improve cargo encapsulation efficiency. For example, incorporating saponin into a microfluidic system increased the loading efficiency of doxorubicin into glioma stem cell–derived sEVs from 19.7 % to 31.9 %, depending on the device configuration and flow rate [130]. Additionally, co-application of saponin with freeze–thaw cycling enabled efficient loading of catalase, resulting in significant neuroprotective effects in a Parkinson's disease model following intranasal delivery [113]. While saponin-assisted loading offers clear benefits, several limitations must be addressed. Saponin can compromise the stability of sensitive biomolecules, potentially diminishing therapeutic efficacy. Moreover, excessive concentrations may cause hemolysis in vivo, posing challenges for clinical translation [129]. Careful optimization of saponin dosage and effective removal of residual surfactant are therefore critical to ensure safety and functional integrity. Key studies employing saponin-assisted loading strategies are summarized in Table 6.

Table 6.

Saponin of engineered sEVs as cargo loading strategies.

Therapeutic cargo	Cell source of sEVs	Concentrations of saponin	Load efficiency	Target diseases	Function	Ref.
Astaxanthin	Fetal bovine serum	0.2 %	>20 %	/	Provide antioxidant and anti-inflammatory protection in HaCaT keratinocytes and RAW264.7 macrophages	[131]
Porphyrins	Endothelial, cancer, and stem cells	0.1 mg/mL	/	/	Achieve an 11-fold increase in hydrophilic-porphyrin loading via saponin assistance versus passive methods	[64]
Catalase	RAW264.7 cells	0.2 %	14.7 %	Parkinson's disease	Deliver catalase with high efficiency and sustained, protease-resistant release, yielding a neuroprotective benefit	[113]
Tyrosinase-related protein-2	Serum	/	43.2 %	/	Enter macrophages and dendritic cells readily, triggering strong TNF-α and IL-6 secretion	[132]
Doxorubicin	SF7761-Glioma cells, U251-Glioma cells	18 μM	31.98 %	Malignant glioma	Curb parent-glioma proliferation while exploiting sEV homing for targeted delivery	[130]

∗Additional recent publications are included in the table, thereby complementing the representative examples discussed in the main text.

2.9. Strategies for loading CRISPR/Cas9 into sEVs: toward targeted gene editing

The CRISPR/Cas system has revolutionized genome engineering by enabling precise, programmable gene editing across diverse cell types. However, safe and effective delivery of components such as sgRNA, Cas9, or dCas9 fusion proteins remains a key challenge. Traditional vectors—including viral systems and lipid nanoparticles—face limitations related to immunogenicity, off-target effects, and cytosolic release inefficiency. Recently, various strategies have been developed to engineer sEVs for CRISPR/Cas delivery, including liposome-sEV fusion, electroporation, focused-ultrasound-guided delivery, RNA-binding protein scaffolds, and surface ligand engineering.

Recent studies have leveraged sEV-liposome hybrid systems to enhance the intracellular delivery and targeting specificity of CRISPR-Cas9 therapeutics. These hybrid nanovesicles combine the biocompatibility and membrane fusion capacity of sEVs with the high cargo-loading efficiency and surface functionalization versatility of liposomes. For instance, CRISPR-Cas9 systems encapsulated within sEV-liposome hybrids have enabled targeted gene editing in MSCs, a cell type typically resistant to transfection [133]. Liang et al. further demonstrated that such hybrid vesicles could selectively deliver CRISPR-Cas9 plasmids to chondrocytes, attenuating cartilage damage [134]. Zhang et al. engineered a CRISPR-Cas9 delivery system using AS1411 aptamer-guided sEV-liposome hybrids (CAELN), enabling nucleolin-targeted suppression of the glutamine synthetase gene (GLUL) in endothelial cell [135]. McAndrews et al. used Exo-Fect to introduce a Cas9/sgRNA plasmid into HEK293T-derived sEVs, yielding the engineered sEVs that selectively edited the oncogenic KRASG12D allele of target cells, reducing its mRNA by 58 % and slowing pancreatic-cancer progression in mice [136].

On the other hand, Kim et al. loaded SKOV3-derived sEVs with CRISPR plasmids by electroporation; the vesicles ablated the poly (ADP-ribose) polymerase-1 (PARP1) locus in ovarian-cancer xenografts and suppressed tumor growth due to cancer tropism-dependent targeting of tumor-derived sEVs [137]. Additionally, Kong et al. employed focused ultrasound to open the blood-brain barrier and deliver RVG-targeted sEVs carrying sgRNA and dCas9-DNMT3A, achieving epigenetic silencing of SNCA in a Parkinson's disease model [138].

Efforts to minimize host responses have shifted Cas9 cargo from plasmid DNA to mRNA or ribonucleoprotein (RNP) formats. Cas9 mRNA can be translated directly in the cytoplasm, thereby averting nuclear entry and reducing immunogenic or oncogenic risk [49]; however, its intrinsic lability shortens intracellular half-life and curtails editing efficiency [139]. Using a CROSS-FIRE reporter assay, de Jong et al. showed that EV-mediated RNA transfer is markedly impaired when key vesicle-biogenesis genes (Alix, Rab27a) are deleted, underscoring the importance of EV sorting pathways for mRNA delivery [140]. Usman et al. addressed mRNA instability by electroporating HA-tagged Cas9 mRNA into red-blood-cell EVs, achieving functional editing in MOLM13 leukaemia and breast-cancer cells [141]. Direct trafficking of the Cas9-sgRNA RNP further mitigates off-target mutagenesis by eliminating transcription and translation in target cells [142,143]. Yao et al. inserted a Com RNA aptamer into the sgRNA loop and expressed a Com-binding peptide at both termini of CD63; the resulting Com/Com interaction enriched Cas9 and adenine-base-editor RNPs within sEVs, yielding ∼0.2 % indel formation in the tibialis anterior muscle of a Duchenne muscular dystrophy mouse model [144]. Post-translational lipidation can also drive RNP loading: Whitley et al. fused an Src-derived myristoylation motif to the N-terminus of Cas9, promoting its incorporation into sEVs [145]. Alternatively, the “NanoMEDIC” system of Gee et al. harnessed an HIV Gag scaffold to co-package Cas9 and sgRNA, achieving >90 % exon skipping in Duchenne muscular dystrophy patient iPSC-derived muscle cells [146]. Collectively, these mRNA and RNP strategies illustrate how precise molecular engineering can endow sEVs with potent, low-toxicity CRISPR payloads suitable for therapeutic genome editing. Representative CRISPR/Cas9-sEV loading strategies are collated in Table 7.

Table 7.

Strategies for loading CRISPR/Cas9 into sEVs.

Therapeutic cargo	Cell source of sEVs	Loading approach	Load efficiency	Target gene	Target diseases/organs/cells	Function	Ref.
CRISPR-Cas9 plasmid	SKOV3/HEK293T cells	Electroporation	79 %	PARP-1	Tumor generated by SKOV3 xenografts	Cancer-derived sEVs selectively accumulate in SKOV3 xenografts and efficiently ferry CRISPR/Cas9 plasmids, enabling in-tumour gene editing	[137]
CRISPR-Cas9 plasmid	HEK293T cells	Transfection of plasmid	/		Hippocampus and cortex	EV-mediated co-delivery of Cre recombinase and Cas9/sgRNA ribonucleoproteins facilitates high-efficiency recombination and genome editing.	[147]
CRISPR-Cas9 plasmid	HEK293T cells	Exo-Fect	/	KrasG12D	Pancreatic cancer	Non-autologous sEVs can encapsulate CRISPR/Cas9 DNA and mediate targeted gene knockout in recipient cells	[136]
CRISPR-Cas9 plasmid	HEK293T cells	Electroporation	31.92 %	MYC	Tumor generated by Raji-bearing xenograft	Anti-CD19-CAR-HEK293T-derived EVs co-display CAR molecules and MYC-targeting CRISPR cargo, conferring tumour-specific tropism and on-cancer editing	[148]
CRISPR-Cas9 plasmid	MSCs	Electroporation	/	TNF-a	Spinal cord	sEV-C@P (hUC-MSC sEVs remodelled with CRISPR plasmid) homes to injury sites, reprograms macrophages and suppresses inflammatory cytokines	[149]
CRISPR-Cas9 mRNA	Red blood cells	Electroporation	50 %	miR-125b	MOLM13 cells	HA-tagged Cas9 mRNA electroporated into red-blood-cell EVs edits leukaemia and breast-cancer cells with minimal toxicity	[141]
CRISPR-Cas9 mRNA	MDA-MB-231 cells	Transfection of Lipofectamine 2000	80 %	mCherry	HEK293T	The CROSS-FIRE reporter showed that deleting Alix or Rab27a markedly reduces EV-mediated RNA delivery	[140]
CRISPR-Cas9 mRNA	HEK293T cells	Transfection of Lipofectamine 2000	/	C/ebpa	Liver	CD9-HuR fusion vesicles enrich miR-155 and deliver it efficiently to recipient cells, down-regulating endogenous targets	[47]
CRISPR-Cas9 RNP	HEK293T cells	Fugene HD (Promega)/polyethylenimine	/	HBB/IL2RG and DMD exon 53	HEK293T-derived HBB-IL2RG EGFP reporter cells	A Com-aptamer/Com-peptide system on CD63 loads Cas9 or ABE RNPs into sEVs, achieving ∼0.2 % indels in dystrophic mouse muscle	[144]
CRISPR-Cas9 RNP	HEK293T cells	Transfection of Lipofectamine 2000	/	stop-DsRed	A549stop-DsRed reporter cell line	GFP-nanobody fused to CD63 captures GFP-tagged Cas9, yielding high-load Cas9 sEVs	[150]
CRISPR-Cas9 RNP	Expi293F cells	Transfection (PEI Max)	/	Fluorescent protein	HEK293T	Light-inducible Cryptochrome 2 dimerisation with CD9—or lipid MPP tagging—drives efficient, opto-controlled Cas9 loading	[151]
CRISPR-Cas9 RNP	HEK293T cells	Transfection (calcium phosphate)	/	Fluorescent protein	HEK293T-eGFP cells	N-terminal Src-myristoylation motif tethers Cas9 into EV membranes for robust export	[145]
CRISPR-Cas9 RNP	Hepatic stellate cells	Electroporation	20 %	PUMA, CcnE1	Liver	sEVs RNP: hepatic-stellate-cell sEVs electroporated with Cas9 RNP accumulate in liver and edit targets in vivo	[72]
CRISPR-Cas9 RNP	HEK293T cells	Transfection of Lipofectamine 2000	/	Duchenne muscular dystrophy	Skeletal muscle cells	NanoMEDIC uses HIV-Gag to co-package Cas9 and sgRNA, achieving >90 % exon skipping in DMD patient-derived myotubes	[146]
CRISPR-Cas9 RNP	HEK293T cells	Transfection of Lipofectamine 2000	/	Duchenne muscular dystrophy	Muscle	Conditional Gag–Cas9 heterodimerisation plus HIV RNA-packaging signals enable ligand-inducible Cas9/sgRNA loading into EVs	[152]
CRISPR-Cas9 RNP	Human or mouse serum	Protein transfectant	/	Duchenne muscular dystrophy	Muscle	Simple electroporation yields SpCas9 RNP-loaded EVs that retain editing activity after delivery	[153]
CRISPR-Cas9 RNP	HEK293T cells	Transfection (EndoFectin Max reagent)	39 %	Fluorescent protein	HeLa cell	Co-expression of CRISPR-Cas9-GFP and GFP-CD63 in 293T cells increases single-GFP-labeled EVs, confirming Cas9 incorporation	[154]
CRISPR-Cas9 RNP	C2C12 cells	Transfection (PEI MAX)	/	miR-29b	Muscle	EVs-Cas9-29b edit miR-29b in vivo, preserving muscle function in immobilisation and denervation atrophy models	[155]
CRISPR-Cas9 RNP	HEK293T cells	Sonication/freeze-thaw cycles	30 %	WNT10B	Tumor	Cholesterol-anchored tetrahedral DNA nanostructures bearing aptamers decorate EVs, steer them to target cells, and deliver Cas9 RNP for GFP or WNT10B knockout	[156]

∗Additional recent publications are included in the table, thereby complementing the representative examples discussed in the main text.

Regarding the selection of optimal cargo-loading methodologies for sEVs necessitates careful consideration of efficiency, biocompatibility, and scalability, which was represented in Table 8. Endogenous loading—achieved through genetic or metabolic engineering of donor cells—offers superior preservation of vesicle integrity and biological function. This strategy is particularly advantageous for delivering large or complex molecular cargos, such as plasmid DNA or multiple CRISPR components, as it harnesses the cell's native machinery for synthesis and sorting [157]. However, its reliance on cellular biosynthesis limits applicability to synthetic compounds and complicates GMP scale-up. In contrast, exogenous methods like electroporation provide a rapid and flexible platform for loading pre-synthesized molecules, most commonly applied to small nucleic acids like sgRNAs or siRNAs, but risk cargo aggregation and membrane disruption [158]. Sonication and extrusion approaches in small-molecule delivery yet compromise sEV structural homogeneity. Passive methods like co-incubation and dialysis are technically accessible but exhibit low efficiency, while saponin-mediated loading enhances hydrophilic cargo uptake at the cost of hemolytic risks. For CRISPR-Cas9 delivery, endogenous fusion with exosome-sorting domains (e.g., CD9-HuR) outperforms electroporation by preserving ribonucleoprotein (RNP) functionality. Collectively, the trade-off between cargo versatility and sEV integrity dictates clinical translation potential, with endogenous strategies favored for biologics and electroporation for nucleic acids. Rigorous quality control (Transmission electron microscopy/Nanoparticle tracking analysis/Western blot) remains non-negotiable to ensure batch consistency.

Table 8.

Comparative analysis of sEVs cargo loading strategies.

Cargo loading strategies	Mechanism	Advantages	Disadvantage	Clinical relevance
Endogenous sEV loading via donor cell modulation	Genetic or metabolic engineering of parental cells.	✅Preserves sEV membrane integrity and natural sorting mechanisms. ✅High biocompatibility. ✅Suitable for nucleic acids/proteins.	❌Limited to biomolecules compatible with cellular machinery. ❌Risk of unintended cargo modification.	➡Ideal for autologous therapies but scalability challenges persist.
Co-incubation-mediated cargo loading into sEVs	Passive diffusion of cargo into isolated sEVs during incubation.	✅Simple, low-cost, and minimally disruptive to sEV structure. ✅Effective for hydrophobic molecules.	❌Low efficiency for hydrophilic/charged cargos. ❌Prolonged incubation may induce aggregation.	➡Best for preclinical proof-of-concept studies.
Electroporation-mediated cargo loading into sEVs	Transient membrane pores induced by electrical pulses.	✅High loading of nucleic acids. ✅Versatile for diverse cargos.	❌Risk of cargo aggregation or sEV lysis. ❌Requires optimization of voltage/pulse duration.	➡Leading method for RNA-based therapies; GMP-compatible protocols emerging.
Sonication-mediated cargo loading into sEVs	Ultrasonic waves disrupt sEV membranes to facilitate cargo entry.	✅High efficiency for small molecules. ✅Rapid process.	❌Potential sEV fragmentation and protein denaturation. ❌Poor reproducibility due to variable energy transfer.	➡Limited to stable cargos; not ideal for biologics.
Freeze-thaw cycle-mediated cargo loading into sEVs	Ice crystals form transient pores during freeze-thaw cycles.	✅Low technical barrier; no specialized equipment. ✅Preserves sEV functionality better than sonication.	❌Low loading efficiency. ❌Repeated cycles cause irreversible aggregation.	➡Supplementary method; rarely used standalone.
Dialysis-mediated cargo loading into sEVs	Hypotonic dialysis swells sEVs, enabling cargo diffusion.	✅Gentle on membrane integrity. ✅Higher loading efficiency than passive incubation for hydrophilic drugs.	❌Time-consuming; risk of protein degradation. ❌Limited to small hydrophilic molecules.	➡Niche applications for sensitive cargos.
Extrusion-mediated cargo loading into sEVs	Mechanical force via porous membranes creates fissures.	✅High loading for chemotherapeutics. ✅Scalable using microfluidic systems.	❌Significant sEV deformation and size heterogeneity. ❌Not suitable for fragile biomolecules.	➡Promising for chemotherapeutic delivery; requires QC refinement.
Saponin-mediated cargo loading into sEVs	Cholesterol-binding surfactant generates membrane pores.	✅Enhances hydrophilic cargo loading. ✅Compatible with complex cargos.	❌Residual saponin causes hemolysis in vivo. ❌Difficult to quantify pore resealing.	➡Requires stringent purification; high immunogenicity risk.
Strategies for loading CRISPR/Cas9 into sEVs	Endogenous: Fusion of Cas9 with EV-sorting domains. Exogenous: Electroporation or streptavidin-biotin conjugation.	✅Endogenous: High functionality; preserves editing activity. ✅Exogenous: Rapid loading; modular design.	❌Low endogenous yield; electroporation may denature RNP complexes. ❌Size exclusion limits Cas9 encapsulation.	➡Endogenous strategies favored for in vivo efficacy.

3. Surface engineering of sEVs for targeted delivery

While sEVs exhibit inherent tropism based on their cellular origin, their native biodistribution is largely confined to clearance organs such as the liver, spleen, and lungs. Consequently, systemically administered sEVs often show limited accumulation at pathological sites, undermining their therapeutic precision.

To address this limitation, a variety of surface engineering strategies have been developed to endow sEVs with targeting specificity toward defined cell types or tissues. These approaches are generally categorized into genetic engineering, chemical modification, and physical methods. More recently, metabolic engineering and immunological modification have also emerged as promising tools for tailoring surface properties to enhance biological recognition and uptake [159,160]. Such modifications, as illustrated in Fig. 7, not only enhance delivery efficiency and therapeutic efficacy but also reduce off-target distribution and systemic toxicity, thereby improving their translational potential.

Fig. 7.

Strategies for surface engineering of sEVs to enhance targeting specificity.

sEVs can be functionally modified on their surface using various bioengineering techniques to improve their targeting ability and therapeutic performance. These strategies are broadly categorized into three main classes: (A) Genetic engineering involves modifying parental cells to express fusion proteins or targeting ligands (e.g., peptides, antibodies, antigens) that are embedded into sEV membranes during biogenesis. Common modifications include LAMP2b-fused peptides (e.g., RVG, GE11), surface-expressed antigens (e.g., HER2), and GPI-anchored nanobodies or receptor-binding proteins. (B) Chemical modification refers to post-isolation covalent or non-covalent conjugation of targeting molecules onto the sEV surface. Approaches include click chemistry (e.g., SPAAC, CuAAC), PEGylation, and electrostatic or hydrophobic interactions using ligands such as aptamers, CD47, or cholesterol-functionalized moieties. These enable the attachment of tumor-homing ligands or immune-evasive signals. (C) Physical methods employ external forces or membrane-manipulating tools such as membrane fusion with targeting vectors (e.g., VSV-G or CRISPR-Cas9 components), magnetic guidance (e.g., SPIONs), laser-responsive particles (e.g., gold nanoshells), or microfluidic membrane engineering to incorporate targeting ligands or create hybrid vesicles. (D) Metabolic engineering leverages the biosynthetic pathways of donor cells to introduce novel functional groups onto the surface of secreted sEVs. Through glycoengineering approaches, such as the overexpression of specific glycosyltransferases or the use of unnatural sugar analogs (e.g., azide-modified sugars like Ac4ManNAz), sEVs can be metabolically labeled in a bio-orthogonal manner. These chemical handles enable subsequent conjugation of targeting ligands or imaging agents via click chemistry, offering a modular and minimally invasive strategy to enhance sEV targeting capabilities. (E) Immunological modification involves decorating the sEV surface with immune-relevant molecules to modulate biodistribution or immune interactions. Strategies include the conjugation of monoclonal antibodies directly onto the sEV surface or the genetic display of immune checkpoint ligands (e.g., CD47) and chemokine receptors (e.g., CXCR4). These modifications can enhance tissue-specific homing, prolong circulation time, and reduce immune clearance, thereby improving the therapeutic efficacy and safety profile of engineered sEVs. Together, these surface engineering strategies enable the customization of sEVs for cell- or tissue-specific delivery, improving biodistribution and minimizing off-target effects. The choice of strategy depends on the cargo type, therapeutic goal, and desired pharmacokinetic profile.

3.1. Genetic engineering for engineered sEVs

Genetic engineering enables the incorporation of targeting ligands onto the sEV surface by modifying donor cells to express fusion proteins comprising transmembrane domains of sEVs (such as Lamp2b and CD63) and homing peptides [161]. This approach allows the generation of sEVs capable of selectively binding to specific cell receptors, thereby enhancing targeting specificity and delivery efficiency [162].

According to our review, lysosome-associated membrane protein 2 (Lamp2b) is one of the most widely utilized scaffolds for surface engineering of sEVs (Table 9). As a member of the lysosome-associated membrane protein family, Lamp2b primarily localizes to lysosomes and endosomes, the major sources of sEVs [31]. Lamp2b consists of a 29-amino acid signal peptide, a large N-terminal extramembrane domain, a transmembrane domain, and a short cytoplasmic tail [163]. Its extracellular domain provides an ideal site for genetic fusion with targeting peptides, enabling the display of specific ligands on the sEV surface [71]. Several studies have demonstrated the versatility of Lamp2b-based engineering strategies [71,87,89,96,100,[164], [165], [166]]. Fusion of a chondrocyte-affinity peptide (CAP) to Lamp2b redirected HEK293T-derived sEVs to articular cartilage, where they efficiently delivered miR-140 and attenuated MMP-13-mediated matrix degradation [96,164]. Swapping CAP for the rabies-virus glycoprotein–derived RVG peptide yielded dendritic-cell sEVs that crossed the blood–brain barrier and silenced neuronal transcripts in vivo [71]. Likewise, iRGD-Lamp2b vesicles demonstrated αvβ3/5-dependent homing to solid tumors and enhanced intratumoral doxorubicin deposition [100], while tLyP-1-Lamp2b sEVs bound neuropilin-1/2 on non-small-cell lung-cancer cells, enabling siRNA-mediated oncogene knock-down [89,165].

Table 9.

Genetic engineering of sEVs as targeting delivery strategies.

sEVs origin	Targeting cells or organs	Homing-molecules	Conjugated molecules	Approaches for homing-molecule loading	Therapeutic mechanism	Ref.
HEK293T cells	Colon cancer cells	Her2	Lamp2b	Transfection	Lamp2-Her2 sEVs: targeting Her2 expressing cancer cells and delivery siFAK1 to inhibit cell proliferation while upregulating the paraptosis level.	[76]
Escherichia coli Nissle 1917	Osteoclasts	DC-STAMP	ClyA	Transfection	FRAT@BEV-DC-STAMP: protect FRAT from enzymatic degradation, and enable its targeted intracellular delivery into pre-osteoclasts.	[83]
Expi293F cells	P21+ CD86+ microglia	Homing peptide CHHSSSARC	Lamp2b	Transfection	Quercetin-EVs: promote recovery and demonstrated safety by mitigating BBB disruption, inducing beneficial microglial polarization, and limiting neutrophil infiltration through its concerted actions.	[171]
Adipose-derived stem cells	M2 microglia	M2 targeting peptide (M2pep)	Lamp2b	Transfection	Lamp2b-M2pep sEVs: Adipose-derived stem cells transfected with a Lamp2b-M2pep fusion secrete sEVs that home to M2-like microglia and suppress ferroptosis via the FXR2/ATF3/SLC7A11 axis	[172]
HEK293T cells	Chondrocyte	Chondrocyte affinity peptide (CAP)	Lamp2b	Transfection	CAP-Nrf2-sEVs: Co-expression of a CAP-Lamp2b construct and Nrf2 in HEK293T cells yields sEVs that block Drp1-Ser616 phosphorylation, prevent mitochondrial fragmentation and protect cartilage	[164]
HEK293T cells	Macrophages	RNA-binding protein HuR	Lamp2b-HuR	Acidification	Lamp2b-HuR sEVs: Endosomal acidification of Lamp2b-HuR sEVs enriches miRNA-binding capacity, resulting in marked anti-fibrotic activity and down-regulation of miR-155 in liver tissue	[46]
Dendritic cells	Synovial fluid-derived MSCs	MSC-binding peptide E7	Lamp2b	Transfection	E7-sEVs: Fusion of the MSC-binding peptide E7 to Lamp2b produces sEVs with high specificity for synovial-fluid MSCs	[71]
Dendritic cells	Tumor cells	av integrin-specific iRGD peptide	Lamp2b	Transfection	iRGD sEVs: Lamp2b-iRGD vesicles preferentially bind αv-integrin-positive breast-cancer cells, enabling doxorubicin delivery and tumor suppression without systemic toxicity	[100]
Dendritic cells	Chondrocyte	CAP	Lamp2b	Transfection	CAP-miR-140 sEVs: CAP-decorated sEVs persist in the joint cavity, penetrate deep cartilage and deliver miR-140, dampening MMPs and slowing osteoarthritis	[96]
Dendritic cells	Chondrocyte	CAP	Lamp2b	Transfection	CAP-Cas9 sEVs: Hybrid CAP-SEVs ferry Cas9/sgMMP-13 plasmids into chondrocytes, knock down MMP-13 and protect cartilage matrix	[134]
HEK293T cells	Liver cancer cells	Apo-A1	CD63 transmembrane protein	Transfection	ApoA1-CD63 sEVs: sEVs presenting ApoA1-CD63 bind SR-BI on HepG2 cells, deliver miR-26a, and inhibit migration and proliferation	[95]
HEK293T cells	Cancer cells	PDL1	PDL1 receptors	Transfection	PD1-Imi sEVs: PD-1 antibody-decorated, imiquimod-loaded sEVs block PD-1/PD-L1 engagement and mature dendritic cells, revitalizing CD8+ T-cell responses	[91]
HEK293T cells	Cancer stem cells	tLyp-1	Lamp2b	Transfection	tLyp-1 siRNA sEVs: Lamp2b–tLyp-1 sEVs knock down oncogenic transcripts and diminish cancer stem cell traits in NSCLC	[89]
Human MSCs	Glioblastoma cells	Angiopep-2/magnetic nanoparticles	Lamp2b/CD63, NHS, EDC, Fe3O4 @mSiO2 NPs	Transfection, NHS/EDC crosslinking chemistry	Magnetic–Angiopep-2 sEVs SPION-loaded, Angiopep-2-modified sEVs cross the BBB under a magnetic field and trigger ferroptosis-based glioblastoma therapy	[87]
HEK293T cells	Severe acute respiratory syndrome coronavirus 2	RBD peptide	ACE2	Transfection	RBD-sEVs: Genetically engineered vesicles displaying the SARS-CoV-2 receptor-binding domain act as coronavirus mimetics for vaccine studies	[173]
MSCs	Neurons	AAV capsid-specific peptides	Lamp2b	Transfection	CAP-Lamp2b sEVs: AAV9-derived CAP fused to Lamp2b enables BBB penetration and neuronal targeting	[174]
Neural stem cells	PDGFRα+OPC	PDGFA-ligand	Virus Lenti-Xstamp-PDGFA	Transfection	TaxI-T3 sEVs: TaxI-peptide sEVs loaded with triiodothyronine promote CNS remyelination in the cuprizone model	[40]
Melanoma B16BL6 tumor cells	Dendritic DC2.4 cells	CpG DNA	Streptavidin and lactadherin fusion protein	Streptavidin-biotin interaction	CpG-SAV sEVs: Streptavidin–lactadherin fusion sEVs present biotin-CpG, potently activating dendritic cells	[175]
Adipose-derived stem cells	Hippocampus (cells expressing a7-nAChR)	Rabies viral glycoprotein (RVG) peptide	RVG-Lamp2b	Transfection	Hippocampus-targeted sEVs: Ligand-engineered vesicles home to the hippocampus and rebalance neuro-inflammation (↓ IL-1α/TNF-α/NF-κB; ↑ IL-10)	[176]
HEK293T cells	Neuronal cells	RVG peptide	Lamp2b	Transfection	RVG-brain-derived neurotrophic factor (BNDP)-sEVs: successfully traverse the blood-brain barrier, targeting neurons in the hippocampus and prefrontal cortex.	[177]
M2 BV2 cells	Brain	RVG29 peptide	Dopamine	/	RVG-neural-target sEVs: rescue functional deficits in stroke by suppressing neuronal apoptosis via upregulation of miRNAs miR-221-3p and miR-423-3p that inhibit p38/ERK signaling.	[178]
BV2 microglia cells	Spine	Ang/RVG peptide	Lamp2b	Transfection	C-A/R-EVs utilize dual-targeting peptides and curcumin pretreatment to reprogram microglia, enhancing myelin debris clearance and suppressing neuroinflammation for synergistic spinal cord repair.	[179]
Expi293F cells	Neuronal	RVG peptide and circMTCO2	Lamp2b	Transfection	The dual-targeting RVG-EVmt system that delivers circMTCO2 to neuronal mitochondria, where it binds ANT1 to inhibit mPTP opening and suppress mtROS release.	[82]
HEK293T cells	Tumor-associated macrophages	CRVLRSGSC	Lamp2b	Transfection	IEEE sEVs: Internally/externally engineered vesicles preferentially enter tumors, silence PI3K-γ and polarize TAMs towards an M1 phenotype.	[180]
Bone marrow MSCs	Cardiomyoblast cell line H9C2	Ischemic myocardium-targeting peptide (IMTP, CSTSMLKAC)	Lamp2	Transfection	IMTP-sEVs: Ischemia-homing peptide (IMTP) sEVs accumulate in infarcted myocardium, reduce fibrosis and apoptosis, and enhance neovascularization	[181]
Murine C2C12 cells	Muscle	CP05	CD63	Amide linker of phosphonodiamidite morpholino oligomer (PMO)	CP05-tagged sEVs: The CD63-binding peptide CP05 enables rapid, universal “painting” of CD63+ sEVs with therapeutic ligands	[169]
Regulatory T cells (Tregs)	Podocytes	RGD peptide	CD63	Transfection	Treg-EVs: target delivery miR-218-5p increase mitophagy in podocytes, reduce cell apoptosis, and alleviate podocyte injury.	[182]
Expi293F cells	CD70-expressing dendritic cells	Glucocorticoid-induced tumor necrosis factor receptor family-related ligand (GITRL)	CD9	A flexible (GGGGS)2 linker	GIFTed-sEVs: CD9 fused to CD70 or GITRL produces vesicles with potent T-cell co-stimulatory activity; a photocleavable CD9 fusion allows light-controlled cargo release	[170]

∗Additional recent publications are included in the table, thereby complementing the representative examples discussed in the main text.

Li et al. engineered a multifunctional nanoplatform (MNP@BQR@ANG‐EXO‐siGPX4) by conjugating anti-CD63-labeled, siGPX4-loaded ANG-Lamp2b exosomes (ANG-EXO) to brequinar (BQR)-loaded Fe₃O₄@mSiO₂ nanoparticles (MNPs). This construct leverages MNPs for brain enrichment, leverages ANG for blood-brain barrier (BBB) transcytosis and GBM targeting, and leverages BQR, siGPX4 and Fe²⁺ for triggers ferroptosis. This synergistic mechanism drives lethal lipid peroxidation, resulting in potent tumor suppression and improved survival in orthotopic GBM models without significant toxicity (Fig. 8) [87]. In addition to Lamp2b, the tetraspanins CD9 and CD63, both highly enriched on sEV membranes, have also been extensively exploited as scaffold proteins for the surface display of targeting ligands [167,168]. Liang et al. engineered tumor-targeted sEVs by stably expressing in HEK293T cells a fusion construct comprising the tetraspanin CD63 and an apolipoprotein-A1 fragment, thereby conferring ApoA1-mediated cancer-cell tropism to the released vesicles [95]. Gao et al. introduced CP05, a short CD63-binding peptide that can be conjugated to diverse cargos, providing a facile, post-isolation method to functionalize or isolate CD63-positive sEVs [169]. Building on the tetraspanin platform, Cheng et al. generated “GIFTed-Exos” by genetically fusing co-stimulatory ligands, CD70 and glucocorticoid-induced tumor necrosis factor receptor family-related ligand (GITRL), to the sEV-associated tetraspanin CD9; the resulting vesicles displayed these molecules on their surface and elicited robust T-cell activation [170].

Fig. 8.

(A) ANG-EXO production and application. An ANG-Lamp2b engineered exosome is electroporation-loaded with siGPX4 and conjugated (anti-CD63) to Fe₃O₄@mSiO₂ magnetic nanoparticles whose mesoporous shell carries the DHODH inhibitor brequinar (BQR). The construct (Fe₃O₄ core ≈20 nm; SiO₂ shell 30–40 nm) is dosed IV with 30 min magnetic localization; ANG (LRP1 ligand) mediates BBB transcytosis and GBM cell uptake, while the acidic endo/lysosome facilitates BQR release (∼68 % at pH 5.5/48 h) and partial exosome–MNP separation. siRNA loading uses ∼1.5 μg siGPX4 per 10⁹ exosomes (typical electroporation input 300 nM), and the composite shows an IC₅₀ ≈ 2 μg mL⁻¹ in LN229 cells. Mechanistically, triple ferroptosis activation is achieved by GPX4 knockdown (siGPX4), DHODH inhibition (BQR), and Fe²⁺ release from Fe₃O₄, driving lipid peroxidation/ROS surges; in orthotopic GBM, this yields superior brain accumulation, tumor suppression, and survival without overt toxicity. (B–C) Uptake of PKH26-labeled exosomes in glioma cells by flow cytometry and confocal imaging. (D) MNP conjugation to exosomes. (E) The mechanism of ferroptosis induction in glioblastoma (GBM) cells. (F) Ex vivo fluorescence images of the major organs upon injection under magnetic field application. Copyright 2022, John Wiley and Sons.

Together, these genetic engineering strategies demonstrate the versatility and modularity of using endogenous sEV membrane proteins, such as Lamp2b, PD1, CD9, and CD63, as scaffolds for targeted ligand presentation. By leveraging the natural trafficking pathways and membrane localization properties of these proteins, engineered sEVs can achieve enhanced targeting specificity through ligand-receptor interaction, improved cargo delivery efficiency, and preserved vesicle integrity. Such advances lay a strong foundation for the development of next-generation sEV-based precision therapeutics.

3.2. Chemical modification for engineered sEVs

Unlike genetic engineering, chemical modification directly attaches targeting ligands onto the surface of sEVs, producing engineered vesicles with enhanced targeting capabilities. By leveraging covalent or non-covalent interactions, this approach enables precise surface functionalization while preserving the structural integrity of the sEV membrane [183]. Recent advances in chemical modification strategies for targeted sEV delivery are summarized in Table 10.

Table 10.

Chemical modification of sEVs as targeting delivery strategies.

sEVs origin	Targeting cells or organs	Homing-molecules	Conjugated molecules	Approaches for homing-molecule loading	Therapeutic mechanism	Ref.
Umbilical cord MSCs	Central nervous system	TAxI peptide	1,2-distearoyl-sn-glycero-3-phosphoethanolaminepoly (ethylene glycol) (DSPE-PEG)	Membrane insertion	TAxI-MSC-EVs exhibited outperformed naïve MSC-Exos in EAE mice: reduced pro-inflammatory T-cell subsets and microglia, promoted M2 polarization (↑ IL-4/IL-10/TGF-β/IDO-1; ↓ IL-2/IL-6/IL-17A/IFN-γ/TNF-α)	[191]
Bone marrow MSCs	Proliferation of pulmonary arterial smooth muscle cells (PASMCs)	Glucuronicc acid (GA)	DSPE-PEG	Membrane insertion	GA-modified sEVs were designed to act as targeting agents for GLUT-1 that overexpressed on the surface of PASMCs, resulting in inhibitory effects on the migration and proliferation of PASMCs and regulation of the gene Camk4 involved in the PAH-associated calcium signaling pathway.	[197]
Bone marrow MSCs	Bone	(AspSerSer)6	DSPE-PEG	Membrane conjugation	α-1,3-fucosyltransferase 6 (Fut6)-(AspSerSer)6-EVs mediated precise bone targeting of curcumin, demonstrating potent anti-osteoporotic effects by restoring BMSC differentiation and mitigating bone loss.	[110]
Expi293F cells	Chondrocyte	Chondrocyte affinity peptide	Sortase A Ligase	Membrane conjugation	Sortase A ligase modified sEVs show ability to prolong joint retention, reduce MMP-13, raise COL2A1, attenuate IL-1β-induced cartilage damage and in OA progression.	[198]
HEK293T cells	Retinal	Photoreceptor-targeting peptide MH42	CP05 fusion peptide	Membrane conjugation	CP05 fusion peptide modified sEVs show elevated repairing efficiency to attenuate retinal dysfunction and photoreceptor loss.	[199]
Umbilical cord MSCs	Neuron	RVG-CP05	TAT-CP05	Membrane conjugation	Dual-modified sEVs with TAT/RVG peptides for targeted delivery of a GPX4 activator to spinal cord lesions, which inhibits ferroptosis and neuroinflammation, promotes neuro-regeneration.	[200]
Milk	Myocardial cells	Ischemic myocardium-targeting peptide (IMTP)	Cholesterol-PEG	Membrane conjugation	Milk-derived sEVs functionalized with a cardiac-targeting peptide to deliver miR30d, which alleviates cardiac hypertrophy and improves function in murine models by targeting GRK5 signaling.	[201]
HEK293T cells	Bone marrow MSCs	EPLQLKM-CHOL(E7-CHOL)	Cholesteryl-anchor	Membrane conjugation	Cholesterol-anchored E7 peptide-functionalized sEVs immobilized on titanium implants via mussel adhesive protein (Exoen-E7@TiM) enable targeted Bmp2 mRNA delivery to BMSCs, bypassing translational suppression via IRES to drive BMP2 expression that enhances osteogenic differentiation and bone-implant integration in rat tibiae.	[202]
M2 Macrophages (RAW264.7)	Neural stem cells (NSC)	IKVAV peptides	Dibenzylcyclooctyne (DBCO)	Click Chemistry	MEXI reprograms macrophages to promote NSC neurogenesis in vitro; homes to the injured spinal cord and improve functional recovery in vivo.	[186]
Umbilical cord MSCs	Activated macrophages	Dextran sulfate (DS)	Metabolic glycan labelling (MGL)-mediated DBCO	Click Chemistry	DS-EVs: target macrophages, reprogramming M1-to-M2 polarization to attenuate osteoclastogenesis and enhance osseointegration via GPC6/Wnt pathway activation.	[203]
Adipose-derived MSCs	Renal tubular epithelial cells	PEGylated hyaluronic acid (P-HA)	DBCO-P-HA	Click Chemistry	Hyaluronic acid-engineered sEVs selectively accumulate in damaged kidneys through CD44/TLR4 interactions after systemic administration.	[204]
Macrophages (RAW264.7)	Subchondral bone	DSS (an optimal ligand for bone targeting)	DSPE-PEG	Membrane insertion	Restored coupled bone remodeling via pSmad2/3-dependent TGF-β inhibition	[192]
Expi293F cells	Chondrocyte	Chondrocyte affinity peptide	DSPE-PEG	Membrane insertion	Target chondrocytes and lowered MMP-13 and raised COL2A1/proteoglycans, protecting cartilage in rat OA.	[93]
iMSCs (derivates of induced pluripotent stem cells	Bone	Bone-targeting-peptide	DSPE-PEG	Membrane insertion	BT-Exo-siShn3 delivered siRNA (Shn3) to osteoblasts, enhanced osteogenesis, inhibited osteoclastogenesis and type-H vessel formation.	[92]
Bone MSCs	Cardiomyocytes	Cardiomyocyte-specific peptide	DSPE-PEG	Membrane insertion	CMP-miR-302-Exo elevates miR-302, Ki67 and YAP in cardiomyocytes, activating Hippo signaling for myocardial repair	[69]
iMSCs (derivates of induced pluripotent stem cells	Cancer cells	Urokinase plasminogen activator (uPA) peptide	DSPE-PEG	Membrane insertion	uPA-sEV-siSrc target senescent stroma/tumor cells; combined with doxorubicin, reduced senescence burden and inhibited xenograft growth.	[68]
Induced neural stem cells (iNSC)	Central nervous system	RVG peptide	PEG-DESP	Membrane insertion	iNSC-EVs treat traumatic brain injury through anti-neuroinflammatory and neuroprotective mechanisms, with efficacy enhanced via localized hydrogel delivery or systemically administered RVG-targeted formulations.	[205]
MDA-MB-231 breast cancer and HT29 colorectal cancer cells	Cancer cells with EpCAM protein	Natural targeting capability	ESD3b and ESD3c with aptamer	Chemical modification	Enhanced CSC targeting; transformed aspirin into an effective anti-CSC agent	[118]
MCF-7 cell line	Cancer cells	Anti-EGFR monoclonal antibodies	Lipid N-dod-PE	EDC/NHS chemistry	Combined sEV fusion and antibody targeting for improved tumor homing	[74]
Bone marrow macrophages	Skeletal stem/progenitor cells (SSPCs)	SSPC-specific aptamers	A Schiff base via aldehyde-amino reaction	Chemical modification	Rescued bone loss in obese mice via precise stromal progenitor targeting	[206]
Osteosarcoma (OS) cells	Osteosarcoma with αvβ3 integrin receptors	Cyclic RGD peptide	DSPE-PEG	Membrane insertion	cRGD-EV-maternally expressed gene 3(MEG3), prepared after elucidating lncRNA MEG3's anti-OS effects, significantly enhances MEG3's anti-osteosarcoma activity through improved tumor-targeting capability.	[44]
HEK293T cells	Chondrocyte	Cationic motifs (cartilage peptide carriers and Avidin)	DSPE-PEG	Membrane insertion	Charge-reversed cationic sEVs deliver RNA via cartilage-targeting arginine-rich motifs anchored in sEVs bilayers using pH-mediated charge-reversal, which can penetrate full-thickness cartilage through weak-reversible GAG ionic interactions, and deliver eGFP mRNA to deep-layer chondrocytes.	[207]
Adipose-derived stem cells	Macrophages (scavenger receptor class A)	DS, consisting of linear 1,6-linked glucopyranoses with sulfate groups	DBCO	Click Chemistry	Engineered sEVs achieve equivalent therapeutic efficacy at one-tenth the dose of bare sEVs, followed by triggering anti-inflammatory cascades through macrophage phenotype modulation.	[188]
bEnd.3 cells	Liver sinusoidal endothelial cell (LSEC)	RLTRKRGLK (RLTR) peptide	DMPE-PEG2000-CRLTRKRGLK	Membrane insertion	RLTR-exosomes, fabricated via DMPE-PEG-CRLTR insertion into bEnd.3-derived vesicles, achieve LSEC-specific enrichment and reduced macrophage clearance through combined RLTR-mediated targeting and CD47 ″don't eat me" signaling.	[208]
Bone marrow MSCs	Chondrocyte	Cartilage affinity peptide (WYRGRL)	Aldehyde-amino reaction	Chemical modification	Cartilage-affinity WYRGRL peptide-modified sEVs can target chondrocytes and significantly inhibit OA-related inflammation and immune gene expression, reversing IL-1β-induced transcriptomic alterations via suppressing catabolism and promoting anabolism.	[209]
Bone marrow MSCs	Alveolar macrophages	Mannose-polyethyleneglycol-N-hydroxy succinimide ester	Amide-Amino binding	Chemical modification	Intratracheally delivered mannose-modified EV-derived miR-23b alleviate ALI by suppressing M1 macrophage activation via the Lpar1–NF-κB pathway.	[78]
Human umbilical vein endothelial cells	Cancer cells (HepG2)	Acetyl galactosamine ammonia, and mannose	DSPE-PEG	Membrane insertion	Tumor cell-derived sEVs were modified with dual ligands (biotin and avidin), which can exhibit targeting ability due to dual ligand targeting according to the mechanism of receptor-mediated endocytosis and anti-cancer effects.	[210]
Bone marrow MSCs	Ischemic brain	c(RDGyK) peptide	DBCO	Click Chemistry	Intravenously administered cRGD-sEVs target ischemic brain lesions; curcumin-loaded variants suppress inflammation and apoptosis in affected regions.	[211]
Neutrophils	Inflammatory synovitis	Ultrasmall Prussian blue (uPB) nanoparticles	DBCO	Click Chemistry	uPB-sEVs selectively target activated fibroblast-like synoviocytes, neutralizing pro-inflammatory factors, scavenging reactive oxygen species, and reducing inflammatory stress. It regulates Th17/Treg cell balance to initiate anti-inflammatory cascades, significantly ameliorating joint damage.	[108]
Blood	Neuron	mAb GAP43	Carbodiimide	EDC/NHS chemistry	Que/mAb GAP43-sEVs specifically targets damaged neurons via mAb GAP43-GAP43 interaction, enhancing neuronal survival through ROS inhibition via Nrf2/HO-1 pathway activation.	[105]

∗Additional recent publications are included in the table, thereby complementing the representative examples discussed in the main text.

Covalent approaches, such as click chemistry and PEGylation, have become essential tools in the field of bioconjugation, providing robust and efficient methods for the conjugation of targeting ligands. Click chemistry, particularly the copper(I)-catalyzed azide-alkyne cycloaddition (CuAAC), is widely recognized for its high specificity, efficiency, and biocompatibility under mild conditions, making it an ideal tool for bioconjugation in therapeutic and diagnostic applications [[184], [185], [186]]. By enabling the precise attachment of diverse functional groups to biomolecules, click chemistry has facilitated the development of advanced targeted delivery systems. For instance, Zeng et al. utilized rapid click chemistry to conjugate IKVAV peptides onto the surface of M2 macrophage-derived sEVs, creating a nano-agent with enhanced homing ability to spinal cord injury sites following systemic administration [186]. Yang et al. developed an acid-cleavable transferrin–decorated, click-engineered exosome platform (ACTE) that co-delivers doxorubicin and siTGF-β to glioblastoma; by TfR-mediated BBB engagement and acid-triggered Tf detachment to enhance transcytosis, ACTE reprogrammed the immunosuppressive milieu and produced marked therapeutic benefit in preclinical GBM models, underscoring the feasibility of precision EV targeting in the CNS [187]. You et al. represented that neutrophil-derived exosomes were engineered into a theranostic platform (uPB-Exo) via copper-free click conjugation with ultrasmall Prussian blue nanozymes. This construct retains both generic exosome and neutrophil-specific membrane proteins, enabling it to actively target inflamed joints through LFA-1–ICAM-1 interactions. Once localized, uPB-Exo exerts potent anti-inflammatory effects by scavenging reactive oxygen species and modulating the Th17/Treg balance, while simultaneously providing T1-weighted MRI contrast for visual monitoring, collectively demonstrating significant efficacy in a collagen-induced arthritis model [108]. Alternatively, a distinct targeting modality was achieved by metabolic glycoengineering of adipose-derived stem cells, yielding exosomes with surface-conjugated dextran sulfate (DS-EXOs). This construct directs the exosomes to pro-inflammatory M1 macrophages via scavenger receptor-A, triggering a therapeutic phenotypic switch to M2 through a combined action of the DS tag and innate miRNAs. This approach highlights a cell-specific immunomodulation strategy that resets the local inflammatory landscape (Fig. 9) [188]. However, concerns regarding the cytotoxicity of copper catalysts have led to the development of strain-promoted azide-alkyne cycloaddition (SPAAC) as a copper-free alternative [189]. In this strategy, azide-modified M1 macrophage-derived sEVs were conjugated to DBCO-modified anti-CD47 and anti-SIRPα antibodies via pH-sensitive benzyl imine bonds. Under the acidic conditions of the tumor microenvironment, these bonds are cleaved, releasing the antibodies and subsequently enhancing tumor cell phagocytosis [190].

Fig. 9.

(A) DS-EXOs as macrophage-reprogramming therapeutic for RA. Metabolic glycoengineering of adipose-derived stem cell introduces azido–sialic acids (Ac4ManNAz, up to 20 μM) on donor-cell glycocalyx, enabling copper-free click attachment of dibenzocyclooctyne–dextran sulfate (DBCO-DS, 10 μM). The DS tag traffics through endosomes into multivesicular bodies and is inherited on secreted exosomes (DS-EXOs). DS engages scavenger receptor-A (SR-A) on activated (M1) macrophages, enhancing joint accumulation after IV dosing and driving M1→M2 polarization; pro-inflammatory cytokines (TNF-α/IL-6/IL-1β) decrease while IL-4/IL-10 increase. miRNA cargo (e.g., let-7b-5p, miR-24-3p) remains unchanged and contributes mechanistically to polarization. (B) Intracellular CD63-GFP and Cy5.5-DS colocalization in MDA-MB-231 cells. (C) Temporal tracking of EEA1 and DS in adipose-derived stem cells. (D) SEM/confocal visualization of DS-decorated EV secretion. (E) SR-A and EXO distribution in inflamed joints. Copyright 2021, The American Association for the Advancement of Science.

PEGylation — the covalent attachment of polyethylene-glycol (PEG) chains to a ligand or lipid anchor — is a staple strategy for prolonging circulation time, increasing hydrodynamic size and shielding immunogenic epitopes. In sEV engineering the workflow is typically two-step: (i) the targeting peptide or protein is covalently grafted to a PEG-lipid such as DSPE-PEG (via NHS-ester, maleimide or click chemistry); (ii) the amphiphilic DSPE tail then embeds non-covalently into the outer leaflet of purified sEVs (post-insertion), displaying the PEG-linked ligand on the vesicle surface. For instance, Wang et al. established TAxI-exos by incorporating DSPE-PEG2000 to enable selective targeting of central nervous system axons in experimental autoimmune encephalomyelitis models [191]. Jing et al. utilized DSPE-PEG to anchor a bone-targeting peptide onto sEVs, restoring coupled bone remodeling in subchondral bone via inhibition of pSmad2/3-dependent TGF-β signaling [192]. Similarly, Zhang et al. conjugated a cysteine-functionalized cartilage affinity peptide (CAP) to DSPE-PEG-MAL, forming CAP-PEG-DSPE, which spontaneously inserted into Expi293F-derived sEV membranes to generate CAP-modified sEVs [93]. Moreover, Gu et al. modified a cardiomyocyte-specific peptide with DSPE-PEG-NHS, allowing covalent attachment to sEVs, followed by electroporation-based loading of miR-302 mimics to enhance therapeutic delivery [69].

Together, these covalent modification strategies offer precise, stable, and biocompatible tools for engineering sEV surfaces with enhanced targeting capabilities. Their modularity enables customizable ligand conjugation while preserving vesicle integrity, making them attractive for translational applications. However, several challenges remain. Copper-based click reactions, while efficient, may pose cytotoxicity risks and require stringent purification, whereas copper-free alternatives like SPAAC, though safer, often involve bulky linkers or lower reaction kinetics. Similarly, PEGylation can hinder cellular uptake or receptor interactions due to steric hindrance if not carefully optimized. Additionally, controlling conjugation site specificity, ligand density, and long-term stability under physiological conditions remains technically demanding. Addressing these limitations is essential for advancing covalently engineered sEVs toward clinical utility.

Non-covalent modifications play a critical role in sEV surface engineering by enabling the reversible and biocompatible attachment of functional ligands without compromising membrane integrity. These interactions, including primarily electrostatic, hydrophobic, and receptor-ligand binding, exploit weak intermolecular forces to anchor therapeutic or targeting moieties onto the sEV surface. Among these, electrostatic interactions are widely employed due to their simplicity and tunability. For example, cationic lipids or peptides can be electrostatically conjugated to negatively charged sEV membranes to enhance cellular uptake or targeting efficiency [193]. Computational tools such as Charger have further advanced our understanding of protein–ligand electrostatics, offering precise quantification of interaction energies and helping optimize ligand design for stable binding [194]. Hydrophobic interactions also facilitate surface modification through the spontaneous insertion of amphiphilic molecules, such as DSPE-PEG–ligand conjugates, into the lipid bilayer of sEVs. This strategy supports ligand display in aqueous environments with minimal membrane disruption, as demonstrated in DNA/lipid interface studies where multivalent hydrophobicity improved structural stability and functionality [195]. Receptor-ligand interactions, often involving a combination of electrostatic and hydrophobic forces, provide high specificity for cell targeting. These interactions have been utilized to guide sEVs to specific tissues or cell types by exploiting native receptor expression patterns. Moreover, bio-inspired polymer systems that use dual-end hydrophobic anchoring further illustrate the potential of non-covalent strategies for constructing stable, multivalent assemblies [196]. Collectively, non-covalent chemical modifications offer a versatile and low-impact approach to engineer sEV surfaces for targeted delivery, although the transient nature of these interactions may pose limitations for in vivo stability and long-term therapeutic retention.

Both covalent and non-covalent surface modification strategies are widely employed to functionalize sEVs for enhanced targeting and therapeutic delivery. Each approach offers distinct advantages and limitations depending on the application context. Covalent modifications provide high stability, long circulation time, and precise control over ligand orientation and density. These methods form irreversible bonds, ensuring that the functional moieties remain stably attached during systemic administration. However, covalent strategies often require harsh reaction conditions, complex synthesis steps, or catalysts (e.g., copper in CuAAC), which may affect cargo bioactivity or sEV integrity. Additionally, scalability and reproducibility under GMP-compliant conditions can be challenging. In contrast, non-covalent modifications rely on reversible interactions, such as electrostatic attraction, hydrophobic insertion, or receptor–ligand binding. These strategies are generally easier to perform, do not require chemical conjugation steps, and preserve membrane integrity. They are particularly useful for modular and rapid ligand screening or temporary functionalization. Nonetheless, the main drawback lies in their lower binding stability, which may result in premature ligand dissociation in vivo, reducing targeting efficacy and therapeutic retention. Therefore, the selection between covalent and non-covalent modification approaches should consider the intended therapeutic context, required circulation stability, regulatory constraints, and production scalability. In some cases, hybrid strategies that combine both methods may offer an optimal balance between stability and biocompatibility.

3.3. Physical methods for engineered sEVs

Physical methods of sEVs have emerged as a versatile and minimally invasive strategy to enhance targeting specificity while preserving vesicle structure and bioactivity. These approaches typically leverage external forces or membrane dynamics to functionalize sEVs, offering simplified protocols and broad applicability. Representative techniques include physical adsorption, magnetic guidance, and membrane fusion, each contributing unique advantages to targeted delivery systems (Table 11).

Table 11.

Physical methods of sEVs as targeting delivery strategies.

sEVs origin	Targeting cells or organs	Homing-molecules	Conjugated molecules	Approaches for homing-molecule loading	Therapeutic mechanism	Ref.
Neutrophils	HGC27 cells/subcutaneous xenograft tumor cells	Superparamagnetic iron oxide nanoparticles (SPION)	Transferrin-transferrin receptor (TfR) interaction	Incubation	SPION-decorated NNVs selectively accumulate at the tumor sites under an external magnetic field, effectively restraining tumor growth and extensively prolonging the survival rate in mice.	[123]
Umbilical-derived MSCs	Kidney	SPION	TfR interaction	Incubation	SPION-EVs magnetically target injured renal sites, delivering CHIP to induce Smad2/3 degradation in tubular cells, thereby alleviating Smad2/3 activation-mediated fibrosis and collagen deposition.	[213]
HEK293T cells	Senescent cells	SPION	TfR interaction	Incubation	Magnetic EV-delivery of iBax mRNA and BAX activator BTSA1 induces apoptosis in senescent cells within atherosclerotic plaques.	[214]
Endothelial progenitor cells	Endothelium	P-selectin glycoprotein ligand-1 (PSGL-1)	Plasma membrane	Extrusion via polycarbonate porous membranes	Neutronphils membrane-modified-sEVs enhance targeting to IRI kidneys via PSGL-1 and promote endothelial repair through miR-21-5p delivery, improving renal function by stimulating EC proliferation, migration and angiogenesis.	[215]
Blood	Mammary carcinoma	C16-K(PpIX)-PKKKRKV	Plasma membrane	Incubation	Dual-stage illumination first employs PCI to rupture endo-lysosomal membranes and release C16-K(PpIX)-PKKKRKV–decorated sEVs into the cytosol; a second light exposure directs porphyrin-mediated photodynamic damage to the plasma membrane and nucleus—via the NLS motif—thereby achieving potent tumor cell ablation.	[216]
Adipose-derived stem cells	Traumatic brain injury	Ultra-small paramagnetic nanoparticles	/	Incubation	engineered sEVs by incorporating ultra-small paramagnetic nanoparticles (USPNs), resulting in sEVs with enhanced miRNA expression and targeted delivery capabilities.	[217]

∗Additional recent publications are included in the table, thereby complementing the representative examples discussed in the main text.

For instance, Jung et al. employed a physical adsorption approach by incubating recombinant IL-2 proteins with T cell-derived sEVs, enabling selective interaction with IL-2 receptor–expressing immune cells without requiring genetic or chemical modification [212]. Zhang et al. developed complementary neutrophil-mimetic nanocarriers using distinct engineering strategies. In the first approach, natural neutrophil-derived exosomes (N-Ex) were magnetically functionalized by incorporating superparamagnetic iron oxide nanoparticles (SPIONs), thereby enabling enhanced targeting under an external magnetic field (Fig. 10A–D). In parallel, serial extrusion was employed to generate neutrophil nanovesicles (NNVs) with substantially improved doxorubicin (Dox) loading efficiency. Subsequent SPION modification of NNV-DOX further augmented their tumor-homing capability. Collectively, this dual-strategy platform synergistically combines the intrinsic biological functions of neutrophil-derived vesicles with the superior drug-loading capacity of synthetic nanovesicles, establishing a versatile and effective system for cancer drug delivery (Fig. 10E–H) [123]. Similarly, Ji et al. developed SPION-labeled mesenchymal stem cell-derived sEVs (SPION-EVs) that overexpressed the carboxyl terminus of Hsc70-interacting protein (CHIP), achieving enhanced renal targeting and antifibrotic effects in kidney injury models under magnetic guidance [213].

Fig. 10.

(A) Schematic of SPION-Ex and exosome-like NVs as a therapeutic and delivery nanoplatform. Neutrophil-derived exosomes (N-Ex) are magnetically functionalized by anchoring transferrin-coated superparamagnetic iron-oxide nanoparticles (SPIONs) onto exosomal TfR—achieved by EDC/NHS conjugation of transferrin to SPIONs, incubation of the resulting Tf-SPIONs with N-Ex for Tf–TfR affinity binding, and magnetic pull-down to remove unbound nanoparticles—yielding SPION-Ex (≈140 ± 11 nm, ζ ≈ −34 mV). In parallel, high-yield neutrophil nanovesicles (NNVs) are fabricated by serial extrusion (11 passes) and loaded with doxorubicin (DOX) (loading 8.5 %), with acid-responsive release (≈74 % at pH 5.5 vs 45 % at pH 7.4, 72 h). (B) Magnetic isolation procedure for SPION-Ex. (C) TEM images of SPION-Ex. (D) Uptake of DiR-SPION-Ex by HGC27 cells with/without MF. (E) NNV-DOX was prepared from DOX-CL-incubated neutrophils, followed by TEM characterization of DOX-CL, blank NNVs, and NNV-DOX. (F, G) Tumor images and sizes in mice following a 24-day treatment with PBS, DOX-CL, NNV, or SPION-NNV-DOX (5 mg DOX/kg) with/without MF. (H) Distribution images of DiR-labeled SPION-NNV-DOX in excised tumors and major organs, acquired 3 days after the last injection. Copyright 2022, The American Association for the Advancement of Science.

3.4. Metabolic engineering for engineered sEVs

Metabolic engineering in sEVs refers to the modification of donor cell biosynthetic pathways using exogenous metabolic substrates (such as unnatural sugars or metabolic analogs), which are incorporated into the surface or cargo of secreted sEVs. This enables precise chemical functionalization of sEV membranes via bio-orthogonal chemistry, enhancing their targeting, stability, or immunomodulatory properties. The details of recent studies related to metabolic engineering on sEVs were represented in Table 12.

Table 12.

Metabolic engineering and immunological modification of sEVs as targeting delivery strategies.

Engineering approach	sEVs origin	Targeting cells or organs	Homing-molecules	Conjugated molecules	Approaches for homing-molecule loading	Therapeutic mechanism	Ref.
Metabolic engineering	HEK-293T cells	Endothelial cells and dendritic cells	Sialyl Lewis X (sLeX) and Lewis X	CD63	Glycoengineering	EV-producing cells co-expressed CD63 extracellular loop-inserted glycosylation domain and fucosyltransferase VII/IX, generating EVs displaying sialyl Lewis X or Lewis X glycan ligands with high specificity toward activated endothelial cells and dendritic cells, respectively.	[221]
Metabolic engineering	PAN02 cells	Tumor	Ac4ManNAz-DP	Azide	Glycoengineering and DBCO	A two-step MGE/SPAAC strategy exploits tumour metabolism for precise targeting: (1) Ac4ManNAz-loaded nanoparticles metabolically install azide groups on tumour cells and their released exosomes; (2) DBCO-functionalized nanoparticles then bind these azide-tagged surfaces via copper-free click chemistry, achieving highly selective tumor accumulation.	[220]
Metabolic engineering	HeLa cells	Cancer	1-(4-carboxybutyl)-4-(7-(4-(diphenylamino)phenyl)benzo[c] [1,2,5] thiadiazol-4-yl)pyridin-1-ium (named TB)	β-D-glucose	Amide-Amino binding	HeLa cells efficiently generate sEVs via a glucose-conjugated photosensitizer (TBG), leveraging glucose transporters and enhanced ATP synthesis for increased biogenesis.	[233]
Immunological modification	CD47-expressing HSC-T6 cells	Activated hepatic stellate cells	5HT1D antibody	CD47 protein	Membrane insertion	sEVs (co-modified with 5HT1D antibody and CD47) enable targeted miR-29b delivery to activated hepatic stellate cells, suppressing TGF-β/SMAD signaling and achieving potent anti-fibrotic efficacy.	[234]
Immunological modification	MC-3T3 cells	Macrophages with expression of CXCL12 (also named SDF-1)	C-X-C chemokine receptor type 4 (CXCR4)	/	Lentivirus	MC-3T3 cells transduced with a CXCR4-expressing lentivirus produce plasma-membrane vesicles enriched in CXCR4. Curcumin can be passively loaded into these CXCR4+ cell-membrane vesicles (CXCR4/Cur-CMVs), creating a homing nanocarrier for SDF-1-riched inflammatory sites.	[226]
Immunological modification	MSCs	SDF-1 high tumor cells	CXCR4	/	Lentivirus	Survivin-targeting siRNA was efficiently loaded into CXCR4-displaying sEVs by electroporation, yielding CXCR4/siSurvivin sEVs for SDF-1–guided RNA delivery to tumor.	[227]
Immunological modification	NIH-3T3 cells	Bone marrow MSCs with expression of SDF-1	CXCR4	/	Transfection	CXCR4-decorated sEV–liposome hybrids loaded with antagomir-188 home to the bone marrow, redirect BMSCs toward osteogenesis, curb adipogenesis, and reverse age-related trabecular loss and cortical porosity in vivo.	[228]
Immunological modification	293T/17 cells	Macrophages with expression of SDF-1	CXCR4	/	Lentivirus	CXCR4-displaying sEVs loaded with miR-126 minimize off-target uptake by macrophages, polarize residual macrophages toward an anti-inflammatory phenotype.	[229]
Immunological modification	Dendritic cells	Hepa1-6-OVA, B16F10, LLC, and MC-38	Anti- PD-1	/	Transfection	Dendritic cells expressing an engineered PD-1 scFv release sEVs bearing surface anti-PD-1, which restore exhausted CD8+ T-cell function and enhance CTL activity.	[231]
Immunological modification	CD8+ T cells	B16F10 cells	PD-1	/	Lentiviral infection	T-cell-derived EVs displaying PD-1 scFv block PD-1/PD-L1 signaling, reactivate CD8+ T-cell proliferation, and enhance tumor clearance.	[235]

∗Additional recent publications are included in the table, thereby complementing the representative examples discussed in the main text.

A well-established method involves culturing donor cells with synthetic monosaccharide analogs like tetraacetylated N-azidoacetyl-D-mannosamine (Ac₄ManNAz), which are metabolically incorporated into glycoproteins and displayed on the plasma membrane and sEV surface. These azide groups can then undergo copper-free click chemistry, such as strain-promoted azide–alkyne cycloaddition (SPAAC), allowing for the covalent attachment of targeting ligands. For example, dextran sulfate (DS)—a ligand for macrophage scavenger receptor-A (SR-A)—has been conjugated to azide-bearing sEVs, resulting in enhanced accumulation at inflamed sites and polarization of macrophages toward an anti-inflammatory M2 phenotype in rheumatoid arthritis models [218,219]. Building on this, Deng et al. developed a two-step bioorthogonal strategy to engineer tumor-derived exosomes (TDEs) as targeted drug carriers. First, donor pancreatic cancer cells were metabolically pre-labeled with azido-sialic acids via in vivo administration of Ac₄ManNAz–PEG–DSPE micelles, resulting in TDEs that naturally display azide groups (Fig. 11A–D). These chemical handles then enabled the gentle, copper-free click conjugation of DBCO-decorated therapeutic micelles, co-loaded with gemcitabine and ISO-1, onto the EV surface (Fig. 11E–F). This "cell-primed, EV-inherited" approach efficiently functionalizes exosomes with drug nanoparticles while preserving their intrinsic membrane proteins and targeting tropism [220].

Fig. 11.

(A) Schematic of experiment design. In this system, tumor-derived exosomes (TDEs) are engineered via donor-cell metabolic glycoengineering: pancreatic ductal adenocarcinoma cells are exposed in vivo to Ac₄ManNAz–PEG–DSPE micelles (IV, once daily × 3), which feed into the sialic-acid biosynthetic pathway to install azido-sialic acids (SiaNAz) on the plasma membrane. Newly secreted EVs inherit these azide handles on their outer leaflet, yielding azide-displaying TDEs (∼100–150 nm; CD9/CD63/TSG101+). Purified TDEs are then post-fabricationally functionalized by copper-free SPAAC click chemistry, wherein DBCO-decorated micelles—co-loading a gemcitabine lipidic prodrug and the MIF inhibitor ISO-1—conjugate to the azide groups on EV surfaces (validated by FRET/TEM co-localization). This “cell-primed, EV-inherited” strategy equips EVs with drug-bearing nanoparticles without harsh surface chemistry, preserving membrane proteins and native tropism while providing programmable attachment site. (B) Flow cytometry quantification of DiD-DBCO-nanoparticle uptake. (C) Co-localization of azide-exosomes (DiI, red) and DBCO-nanoparticles (CFPE, green). (D) TEM of azide-exosome and DBCO-nanoparticle interaction. (E) FRET mechanism between DBCO-nanoparticles and azide-exosomes. (F) FRET detection spectrophotometrically; controls: non-DBCO nanoparticles and PBS. Copyright 2023, John Wiley and Sons.

Additionally, a modular glycoengineering strategy was reported that combines: (a) a display-compatible EV scaffold protein, (b) a glycosylation domain for ligand presentation, and (c) a glycosyltransferase for site-specific glycan synthesis [221]. Supporting this concept, Williams et al. demonstrated that enzymatic trimming of surface glycans significantly altered EV uptake and biodistribution, underscoring the functional relevance of glycan composition in vesicle-cell interactions [222].

3.5. Immunological modification for engineered sEVs

A broad body of evidence indicates that sEVs, particularly those derived from MSCs, exert potent immunomodulation across autoimmune and inflammatory models [223]. Immunological modification of sEVs refers to the strategy of engineering vesicles with immune-related molecules, such as antibodies, immune checkpoint regulators, cytokine receptors, or chemokine receptors, to modulate immune interactions, evade immune clearance, or enhance targeting to immune-enriched microenvironments. Immunological modification has emerged as a pivotal strategy for enhancing the targeted delivery of sEVs in cancer therapy.

One widely adopted approach is the anchoring of immune-modulatory or targeting antibodies on the sEV surface. For example, CD47-modified sEVs interact with signal regulatory protein alpha (SIRPα) on macrophages to inhibit phagocytosis, thereby prolonging circulation time and enhancing accumulation at EGFR-positive triple-negative breast cancer (TNBC) lesions [224]. In another example, SAV-CpG engineered sEVs, functionalized with streptavidin-lactadherin fusion proteins, enable the precise coupling of biotinylated CpG DNA to the sEV membrane, enhancing antigen presentation and stimulating antitumor immune responses [225].

Another immunological strategy centers on the CXCR4-SDF-1 axis, which governs the homing of immune and stromal cells to inflamed or tumor tissues [[226], [227], [228]]. Several groups have genetically engineered donor cells to overexpress CXCR4, generating sEVs that selectively migrate to SDF-1-rich microenvironments. These CXCR4-enriched sEVs have demonstrated targeted delivery to inflamed colonic tissue [226], tumor sites [227], and the bone marrow niche [228]. For instance, Hu et al. reported that CXCR4+ sEVs derived from NIH-3T3 cells preferentially accumulated in the bone marrow due to high SDF-1 levels, enabling hematopoietic stem cell targeting [228]. Similarly, Luo et al. utilized CXCR4-overexpressing sEVs loaded with miR-126 to reduce macrophage uptake and promote anti-inflammatory polarization [229].

Besides, the common target for immune checkpoint inhibition, programmed cell death ligand 1 (PD-L1), which is overexpressed in many cancer types and binds the immune inhibitory molecule programmed cell death protein (PD-1), was investigated as a target for sEVs delivery. Wiklander et al. loan engineered a universal EV targeting platform by screening fusion constructs to identify CD63–z as the optimal scaffold for displaying Fc-binding domains on EVs (Fig. 12A–B). These "Fc-EVs" can be decorated with any therapeutic antibody, such as HER2 and PD-L1 antibodies, achieving a dramatic >500-fold increase in target cell uptake - e.g., raising EV-positive cell fraction from 11 % to 76 % (Fig. 12C–E). When further loaded with doxorubicin, the antibody-guided Fc-EVs synergize to enable potent tumor control and survival benefit, effectively merging antibody specificity with EV delivery in a single, tunable nanocarrier [230]. Liu C. et al. developed a novel ASPIRE (Antigen Self-Presentation and Immune-checkpoint Reversal) platform in which dendritic-cell sEVs are engineered to co-display tumor neoantigens and membrane-anchored anti-PD-1 antibodies [231]. By integrating antigen presentation with local checkpoint blockade, this design simultaneously primes CD8⁺ T cells and relieves PD-1-mediated suppression, thereby overcoming immune tolerance and amplifying cytotoxic T-cell responses.

Fig. 12.

(A) Schematic of engineered producer cells generating Fc-binding domain-displaying EVs (Fc-EVs) for universal antibody conjugation, therapeutic loading, and tissue-specific targeting. Producer cells are engineered with fusion constructs that couple an Fc-binding domain to an EV-sorting scaffold (systematically screened across 9 Fc binders × 9 sorting domains); CD63–z emerged as the optimal combination, yielding the highest expression and single-vesicle antibody-binding efficiency, so-called “Fc-EVs.” Antibody decoration dramatically increases cell-specific uptake—for example, PD-L1 antibody (atezolizumab) boosts Fc-EV internalization 509-fold in IFNγ-stimulated B16F10 cells and raises the fraction of EV-positive target cells from ∼11 % to ∼76 %; HER2 targeting shows similar gains in SKBR-3 cells. The platform supports chemotherapeutic loading by electroporation (e.g., 0.34 μg doxorubicin per 1 × 10¹¹ EVs), converting modest single-agent effects into significant tumor control and survival benefit when combined with targeting antibody. Fc-EVs bridge the clinical maturity of monoclonal antibodies with the delivery advantages of EVs, providing a universal, plug-and-play targeting layer that is interchangeable across indications while preserving nanoscale size and biocompatibility. The result is a flexible, low-dose–enabled, antibody-guided EV drug carrier with validated tumor tropism and therapeutic efficacy, well-suited for translation and adaptable to diverse antibody–cargo pairings. (B) Screening of EV-sorting and Fc-binding domain combinations to optimize Fc-EV design. (C) Uptake of mNG + Fc-EVs by HER2+ SKBR-3 cells with/without trastuzumab by flow cytometry. (D) Uptake of atezolizumab-decorated Fc-EVs in IFNγ-stimulated/unstimulated B16F10 cells. (E) Confocal imaging confirming enhanced Fc-EV internalization with PD-L1-Ab. Copyright 2024, Springer Nature.

While these immunological modifications significantly improve targeting performance, they also pose challenges. The introduction of exogenous immune proteins may increase immunogenicity, and repetitive administration could lead to unintended immune modulation or suppression. Current efforts to address these risks include glycan masking, hybrid membrane coatings, and humanized antibody engineering to balance immune compatibility with therapeutic efficacy [232]. Taken together, immunological engineering enables sEVs to serve not only as passive carriers but also as active immune modulators and precision delivery vehicles. Continued refinement of these techniques will be critical for advancing their clinical translation in oncology, inflammation, and regenerative medicine. The details of recent studies related to immunological engineering of sEVs were represented in Table 12. Table 13 systematically summarizes the respective advantages and limitations of contemporary surface-engineering strategies used to confer targeted-delivery capability on sEVs.

Table 13.

Comparative analysis of surface engineering strategies for targeted sEVs drug delivery.

Target delivery strategies	Mechanism	Advantages	Disadvantage	Clinical relevance
Genetic engineering	Fusion of targeting ligands (e.g., RVG, iRGD) with exosomal membrane proteins (e.g., Lamp2b, CD9) via viral/non-viral transfection of donor cells.	✅High specificity. ✅Endogenous display of preserving membrane integrity and reducing immunogenicity.	❌Limited to peptide ligands. ❌Complex protein folding not guaranteed. ❌Viral vector risks.	GMP-compatible stable cell lines available.
Chemical modification	Covalent conjugation or non-covalent insertion into isolated sEVs.	✅Modularity: Co-display of targeting ligands, PEG, and fluorescent dyes via click chemistry. ✅Broad ligand compatibility: Antibodies, aptamers, synthetic polymers.	❌Potential membrane damage. ❌Batch heterogeneity. ❌Requires post-reaction purification.	Key for antibody-directed tumor targeting.
Physical methods	Magnetic guidance: SPION-loaded sEVs steered by external fields.	✅Real-time control of magnetic fields. ✅No chemical residues.	❌SPIONs may induce oxidative stress. ❌Low fusion efficiency.	Emerging for deep-tissue targeting.
	Membrane fusion: Hybrid sEV-liposomes with targeting ligands.
Metabolic engineering	Incorporation of bioorthogonal handles into donor cells for click chemistry-mediated ligand conjugation.	✅Site-specific labeling: Azide-modified sEVs enable covalent ligand attachment with high efficiency. ✅Native-like membranes: Maintain "self" markers (CD47) to evade clearance.	❌Metabolic toxicity of non-natural sugars/amino acids. ❌Scalability challenges in bioreactors.	Ideal for personalized oncology.
Immunological modification	Display of immune-modulating molecules (e.g., PD-L1, CD47) or antibody fragments.	✅Dual functionality. ✅Enhanced circulation.	❌Risk of cytokine storms or autoimmune reactions. ❌Antibody fragments may dissociate in serum.	Frontrunner for combination immunotherapy.

4. Endosomal escape strategies for engineered sEVs

Following cellular uptake, therapeutic sEVs are routed through the canonical endo-lysosomal pathway: vesicles are first internalized into early endosomes (EE), mature into late endosomes (LE), and ultimately accumulate in acidic lysosomes (pH ≈ 5.0). The progressive drop in pH and the action of hydrolytic enzymes can inactivate or degrade encapsulated molecules. Although naïve sEVs display superior endosomal escape relative to synthetic lipid nanoparticles—an advantage often attributed to their distinctive membrane lipid composition and associated proteins [236]—additional engineering is usually required to achieve robust cytosolic delivery.

Engineered sEVs and related nanocarriers exploit four canonical biophysical strategies to breach the endosomal barrier and deliver cargo to the cytosol. (i) The proton-sponge effect relies on pH-buffering polymers—classically polyethyleneimine (PEI) or pH-responsive polymer vesicles—that sequester protons in the maturing endosome, drive compensatory Cl⁻/water influx, and trigger osmotic swelling until the membrane ruptures. (ii) In disassembly-and-insertion, dual pH-/redox-sensitive polymer vesicles fragment into amphiphilic monomers under acidic or reductive cues; these monomers intercalate into the endosomal bilayer, destabilizing and lysing it. (iii) A volume-expansion stress mechanism harnesses nanoparticles that swell or undergo a lamellar-to-hexagonal phase transition upon protonation, generating mechanical tension that tears the endosomal membrane—often synergistic with proton-sponge chemistry. (iv) Finally, membrane fusion (“back-fusion”) is achieved by incorporating fusogenic lipids or viral envelope proteins such as VSV-G onto the vesicle surface; once endocytosed, these proteins catalyze direct fusion between the nanovesicle and endosomal membranes, dumping the cargo directly into the cytosol. Individually or in concert, these mechanisms convert sEVs from passive carriers into active, virus-mimetic platforms capable of efficient intracellular drug delivery [237].

Current endosomal-release technologies converge on five mechanistic classes:

• 1)Endosomolytic Peptides: A promising strategy to optimize small extracellular vesicle (sEV) therapeutics involves functionalizing their membranes with pH-dependent fusogenic peptides, such INF7 or GALA, derived from influenza hemagglutinin. These peptides, anchored using cationic lipids, facilitate enhanced cellular uptake and promote endosomal membrane fusion or pore formation under acidic conditions. Following endocytosis, the peptides integrate into the endosomal membrane, disrupting its integrity and enabling an efficient release of sEV cargo into the cytoplasm [[238], [239], [240]].
• 2)Fusogenic proteins: Incorporating viral or bacterial pore-forming glycoproteins onto sEV membranes provides an active, bacteria OR virus-mimetic route for endosomal escape. The vesicular stomatitis virus glycoprotein VSV-G confers a “back-fusion” mechanism: once endocytosed, VSV-G mediates direct fusion between the sEV and endosomal membranes, releasing cargo into the cytosol and bypassing lysosomal degradation. VSV-G-decorated sEVs have achieved functional protein delivery (e.g., Cre recombinase) to >90 % of target cells—an order-of-magnitude improvement over naïve vesicle [147]. A complementary strategy employs Listeriolysin O (LLO) from Listeria monocytogenes; at acidic endosomal pH, LLO forms transient pores that likewise permit rapid cytosolic release of nanoparticle payloads [241]. These peptides represent potent tools for enhancing intracellular delivery via engineered sEVs.
• 3)pH-responsive materials: Endosomal acidification (pH ≈ 5–6) can be exploited by coating or hybridizing sEVs with acid-sensitive lipids or polymers that remain inert at pH 7.4 yet become membrane-disruptive in the endosome. (i) Lipid hybrids: Incorporating fusogenic “cone-shape” lipids that undergo a lamellar-to-hexagonal phase transition at low pH destabilizes the endosomal bilayer, enabling rapid cytosolic release of RNA and protein cargo [242] (ii) Proton-sponge polymers: Cationic polymers (e.g., PEI, histidine-rich polypeptides). grafted onto sequester protons, drive chloride influx, and cause osmotic swelling that ruptures the endosome [243]. Both designs have produced sEV-liposome or sEV-polymer hybrids that deliver nucleic acids and enzymes far more efficiently than unmodified vesicles, highlighting pH-responsive materials as a versatile toolkit for endosomal escape.
• 4)
  Light-Triggered Release Systems: Emerging “on-demand” strategies use external stimuli like light to induce endosomal escape of sEV cargos at desired times and locations. Optically responsive nanovesicles are redefining control over the intracellular fate of sEV cargo. Two complementary paradigms have emerged.

  • a.
    Photochemical internalization (PCI). Porphyrin- or phthalocyanine-based photosensitizers can be loaded into or conjugated onto sEV membranes. Upon low-intensity light exposure, the photosensitizer generates singlet oxygen, peroxidising the endosomal bilayer and spilling the vesicular payload into the cytosol with sub-micron precision. In a dual-stage design, chimeric-peptide sEVs first use PCI to breach the endosome and subsequently execute nucleus-targeted photodynamic therapy, eradicating tumor cells in vivo [216]. Macrophage-derived Zn-phthalocyanine sEVs similarly combine efficient PCI with immunogenic cell death, eliciting systemic anti-tumor immunity [244].
  • b.
    Photothermal activation. NIR-absorbing coatings-melanin-mimetic polydopamine, hollow gold nanostructures, or aptamer-tethered gold nanorods—convert light to heat, creating a local burst that ruptures endosomal membranes or directly ablates tumor tissue. Polydopamine-sheathed, magnetically guided sEVs released doxorubicin and an anti-miR-21 beacon on command, shrinking orthotopic breast tumors by >97 % in mice [245]. Stem-cell sEVs loaded with hollow gold nanoparticles accumulated in primary and metastatic lesions; short NIR pulses produced hyperthermia that eradicated multi‐nodular pancreatic tumors in vivo [246]⁴, while aptamer-AuNR hybrids achieved selective photothermal killing of cancer cells in vitro [247].

Collectively, these PCI and photothermal platforms deliver spatiotemporally gated, endosome-escaping therapeutics, transforming sEVs from passive shuttles into precisely activated nanodevices with superior efficacy and minimal off-target toxicity.

• 5)Pharmacological adjuvants: In addition to protein, peptide, and materials engineering, several small-molecule adjuvants might be co-administered to boost sEV uptake or promote the endosomal release of their cargo. Amphotericin B, a polyene that intercalates into lipid bilayers, transiently increases plasma-membrane permeability and thereby augments vesicle internalization [248]. Once inside the endosomal pathway, weak bases such as chloroquine, amantadine, and NH₄Cl accumulate and become protonated; the ensuing “proton-sponge” or pH-buffering effect drives Cl⁻ and water influx, causes endosomal swelling and rupture, and markedly boosts the cytosolic release of nucleic acids and proteins [[249], [250], [251]]. In contrast, bafilomycin A₁ inhibits the vacuolar H⁺-ATPase, preventing endosomal acidification and maturation into degradative lysosomes, thereby prolonging the window for cargo escape [252]. UNC10217938A—identified in a phenotypic screen—directly destabilizes endosomal membranes and delivers a >10-fold increase in functional antisense-oligonucleotide activity [253]. Uptake can also be amplified at the plasma-membrane level: the iTOP formulation promotes macropinocytosis and facilitates protein release from macropinosomes [254], while epidermal growth factor (EGF) activates EGFR-driven macropinocytosis in oncogenic KRAS cells, selectively enhancing EV internalization [255]. Collectively, these pharmacological boosters provide a versatile adjunct to vesicle engineering strategies, either by increasing the number of sEVs that reach the endosomal compartment or by weakening the endosomal barrier itself, thus maximizing the therapeutic payload that reaches the cytosol.

Endosomal escape remains the principal intracellular bottleneck for sEV therapeutics. By integrating (i) stimulus-responsive polymers and lipids, (ii) viral- or toxin-derived fusogenic proteins, (iii) pH-activated lytic peptides, (iv) optically triggered photosensitizers or photothermal agents, and (v) pharmacological adjuvants that enhance uptake and membrane disruption, researchers have transformed sEVs from passive vesicles into virus-mimetic nanodevices capable of precision cytosolic delivery. These orthogonal strategies can be combined, e.g., magnetic targeting plus NIR photothermal burst or proton-sponge polymers with fusogenic peptides—to achieve synergistic release profiles while minimizing off-target toxicity. Looking forward, rational selection and modular assembly of these escape modules, guided by quantitative endosomal-escape assays and in vivo imaging, should enable “plug-and-play” sEV platforms that match or surpass viral efficiency yet retain the safety and immunological advantages of natural vesicles.

5. Artificial intelligence-driven future applications in engineered sEVs

Artificial intelligence (AI) is increasingly reshaping the biomedical landscape by enabling the dissection of complex biological systems and the optimization of therapeutic strategies. While current applications of AI to sEVs have predominantly focused on diagnostic assay development, broader applications are beginning to emerge. The convergence of AI with vesicle engineering is expected to transform targeted drug delivery, rational cargo selection, and large-scale manufacturing, thereby advancing the clinical translation of engineered sEVs. Nevertheless, this convergence remains at an early stage of development, and significant efforts are required to establish robust frameworks, curated datasets, and validated pipelines before full translational potential can be realized.

• 1)AI-guided design of precision targeted sEV delivery

The rational design of sEVs for targeted delivery can be enhanced by AI-driven prediction of ligand–receptor interactions. Computational peptide design [256,257] and glycoengineering approaches [258] have produced high-affinity motifs that can be incorporated into vesicle membranes to improve tissue tropism. Machine learning models have the capacity to accelerate this process by ranking ligand binding affinities and minimizing off-target interactions. Cargo design also benefits from AI, as predictive algorithms can guide the selection of nucleic acid or protein payloads that maximize therapeutic potency while preserving vesicle stability [259,260]. In addition, AI may uncover novel bioactive cargo molecules or mechanisms of vesicle uptake, providing new strategies for constructing advanced drug delivery systems. Early investigations into AI-enabled nucleic acid delivery underscore the importance of parameters such as hydrophobicity, size, and surface potential, which, if systematically applied to EVs, could substantially refine delivery efficiency in the future [261]. Despite this promise, the availability of curated datasets covering APIs, excipients, and known formulations against biological targets remains limited, constraining the broader integration of AI into formulation and delivery development.

• 2)AI-enhanced manufacturing and quality control

The clinical translation of engineered sEVs requires scalable manufacturing processes supported by rigorous quality control. AI-enabled multiplexed vesicle analysis provides quantitative benchmarks for product characterization and diagnostic relevance. Automated image analysis based on convolutional neural networks allows label-free detection of vesicle morphology and apoptotic contaminants during production, while machine learning applied to nano-flow cytometry discriminates vesicle subpopulations and supports potency assays for batch release [262,263]. Together, these technologies align with process analytical technology principles in biologics, reinforcing reproducibility and GMP compliance. In future applications, such AI-assisted frameworks could be extended to fully automated EV production workflows—including cell culture, medium extraction, pre-processing, vesicle isolation, and characterization of size distribution and total RNA/protein content. Modulating culture conditions with defined environmental factors or molecular additives could be systematically modeled and optimized, ultimately yielding standardized, high-quality therapeutic EV products.

6. Clinical translation and perspectives

The clinical translation of engineered sEVs is poised to redefine precision therapeutics by enabling targeted delivery with minimized off-target exposure. Advances in cargo-loading (RNA, protein, and small molecules) and surface engineering (ligand display and membrane reprogramming) have accelerated entry of engineered sEVs into early-phase trials across multiple indications, underscoring their broad therapeutic promise. Yet, large-scale adoption remains constrained by persistent challenges. Source heterogeneity—spanning autologous, allogeneic, and even plant-derived vesicles—drives variance in cargo composition, surface epitopes, and immunogenic potential, reinforcing the need for harmonized, GMP-compliant manufacturing and release criteria. Moreover, while genetic and chemical modifications can sharpen tissue tropism, these alterations may elicit unintended immune responses in repeat-dose settings, particularly over chronic administration [264]. Recent reviews delineate actionable levers to increase sEV yield—optimizing producer-cell biology (e.g., 3D/perfusion culture, hypoxic or nutrient conditioning), applying controlled mechanical cues (shear/bioreactor agitation), and pharmacologically modulating EV biogenesis and release—while emphasizing that any throughput gain must be balanced against vesicle integrity and phase-appropriate critical quality attributes (CQAs) [33]. From a process standpoint, prevailing downstream platforms still impose a yield–integrity trade-off: ultracentrifugation and tangential-flow filtration are scalable but can perturb membrane architecture, whereas microfluidic systems better preserve vesicle features yet remain cost-intensive for industrial deployment [265].

From a Chemistry, Manufacturing, and Controls (CMC) perspective, engineered sEVs add layers of complexity relative to naïve products. Achieving batch-to-batch consistency is intrinsically difficult due to donor-cell biology and the diversity of engineering workflows. Conventional QC modalities—nanoparticle tracking analysis, Western blotting, ELISA—lack the granularity to discriminate naïve versus engineered vesicles, to quantify cargo-loading efficiency, or to measure surface ligand density with the fidelity required for lot release. Regulatory expectations for potency assays and for defining CQAs are still evolving, further complicating release specifications. In response, next-generation analytical tools are reshaping characterization and QC. Nano-flow cytometry (NanoFCM) enables single-vesicle resolution of size, concentration, and surface markers; single-particle interferometric reflectance imaging (SP-IRIS) supports label-free, multiplexed protein detection; tunable resistive pulse sensing (TRPS) refines particle sizing and concentration; and microfluidic immunocapture platforms (e.g., ExoView) integrate phenotyping with semi-quantitative assessment of engineered ligand density. Complementary omics (mass-spectrometry proteomics and RNA-seq) provide high-precision cargo profiling. For targeted designs, surface plasmon resonance (SPR) and biolayer interferometry (BLI) offer label-free kinetics and affinity/avidity measurements under near-physiologic conditions, while fluorescence-based ligand–receptor binding assays (aptamer/antibody based) verify specificity and epitope integrity post-fabrication. Despite these gains, the proportion of functionally engineered vesicles—that is, particles that both carry the intended payload and display the intended targeting moieties—often remains modest, leaving a substantial fraction of non-functional or partially modified sEVs that dilute apparent potency and complicate purification and quantitative release testing. These limitations argue for strengthened process analytical technologies (PATs) and enrichment strategies to boost the yield of bioactive vesicles. Collectively, these insights argue for process-analytical-technology–guided optimization that couples upstream intensification with integrity-preserving purification to maintain lot-to-lot comparability.

Looking forward (Fig. 13), convergent innovations are likely to overcome current bottlenecks. AI-guided peptide and protein design can yield sEV-mimetic ligands with optimized binding kinetics and tissue selectivity, improving uptake and endosomal escape. Orthogonally, the coupling of targeting sEVs with CRISPR–Cas platforms is enabling virus-free, one-shot gene modulation, advancing toward customizable, patient-specific interventions with reduced off-target liabilities. Hybrid engineering that layers modular surface chemistry, bioorthogonal click reactions, and nucleic-acid nanostructures could support programmable control over both cargo and targeting, tunable to disease-specific molecular signatures. Such systems open avenues for next-generation sEVs capable of real-time therapeutic responsiveness, immune evasion, and microenvironmental sensing.

Fig. 13.

Translational landscape of engineered sEVs: current hurdles and emerging solutions.

Realizing these capabilities in the clinic requires a co-evolving regulatory science. Clear, harmonized standards for identity, purity, potency, and safety are essential, including consensus CQAs, validated potency assays linked to mechanism of action, and requirements for long-term immunogenicity, biodistribution, and shedding/clearance assessments. Close coordination among bioengineers, clinicians, and regulators will be critical to ensure that innovation in sEV design is matched by manufacturability and regulatory readiness.

Finally, we envisage a vaccine-like, in vivo sEV therapeutic platform that directly addresses dominant CMC and cost constraints of ex vivo manufacture. In this paradigm, engineered nucleic-acid constructs are delivered to disease-relevant (or strategically selected non-disease) tissues so that host cells transiently assemble and secrete targeted, cargo-defined sEVs in situ (or ex situ for endocrine-like action). The constructs (i) encode therapeutic cargoes and membrane display modules for tissue targeting, (ii) incorporate genetic elements that enhance EV biogenesis and cargo loading, and (iii) embed drug-inducible switches to titrate production and define treatment windows. Pharmacologic modulators of EV release provide a second, orthogonal dial on secretion rate, while liquid-biopsy EV analytics (surface epitopes and cargo surrogates) and standard pharmacodynamic biomarkers support real-time monitoring of output and effect. By eliminating large-scale cell culture, purification, and lot-release testing, this strategy reduces CMC burden and costs; by shifting production to programmable host cells, it dampens batch variability; by generating vesicles within pathological microenvironments, it improves spatial pharmacology; and by combining genetic switches with secretion-level tuning, it enables precise, repeatable dosing. Collectively, this in vivo–generation framework offers a scalable, controllable, and clinically monitorable route for engineered sEV therapeutics that aligns with translational, manufacturing, and regulatory imperatives—while preserving the flexibility needed to adapt to disease-specific design rules.

In summary, engineered sEVs are nearing an inflection point: their success will depend on closing the loop between mechanism-specific potency, real-time quality control, and economically viable CMC. We argue that a biomarker-driven development path—augmented by PATs and, where appropriate, a vaccine-like in vivo generation strategy—offers a scalable, controllable, and regulator-ready route to broad clinical impact. This integrated framework preserves design flexibility for indication-specific needs while directly addressing the cost and standardization barriers that have constrained the field.

The red inner hub depicts the principal obstacles that impede clinical deployment of engineered sEVs. Biological challenges include immunogenicity, off-target effects and rapid systemic clearance. Production constraints encompass vesicle heterogeneity and the absence of scalable GMP workflows. Regulatory and commercial barriers arise from immature guidance documents and unresolved intellectual property issues. Technical limitations—low cargo-loading efficiency, vesicle instability and incomplete targeting—compound these difficulties, while standardization gaps in purification and contamination control complicate batch release. The surrounding green ring summarizes multidisciplinary innovations aimed at each bottleneck: (1) AI-driven target selection and sequence optimization streamlines identification of disease-relevant genes and optimizes guide-RNA or peptide ligands. (2) Modification and sequencing of effective vectors integrates biorthogonal chemistry and nucleic-acid nanotechnology to confer endosomal escape, degradation resistance and receptor-specific homing. (3) Engineering modifications for targeted delivery exploit modular surface ligands, hybrid membranes and micro-environment-responsive cargos. (4) Standardization introduces refined purification schemes and contamination controls, supported by advanced analytics (NanoFCM, SP-IRIS, TRPS, SPR/BLI). (5) Production innovations focus on scalable bioreactors, microfluidic manufacturing and real-time process-analytical technologies. The lower panel illustrates the bench-to-bedside continuum: fundamental target discovery and compound screening feed into in vitro/in vivo pre-clinical studies; iterative feedback informs Phase I/II/III clinical trials, dosage optimization and safety monitoring; translational approval completes the loop, guiding next-generation modifications, structures and delivery platforms.

CRediT authorship contribution statement

Hongtao Xu: Writing – original draft, Investigation, Funding acquisition, Formal analysis. Rui Liu: Writing – original draft, Investigation, Formal analysis, Conceptualization. Hao Zhou: Writing – original draft, Investigation, Formal analysis, Conceptualization. Bin Kong: Methodology, Investigation. Kai Shen: Methodology, Investigation. Tao Zhao: Validation, Funding acquisition. Xiaofeng Du: Validation. Hao Zhang: Validation. Huanghe Song: Validation. Dunming Guo: Validation. Xiaoyuan Gu: Validation. Qing Wang: Validation. Chien-Wei Lee: Writing – review & editing, Supervision, Funding acquisition. Guoyong Yin: Supervision. Yingze Zhang: Supervision. Wei Chen: Writing – review & editing, Supervision, Funding acquisition, Conceptualization.

Ethics approval and consent to participate

This article is a review paper and does not contain any studies with human participants or animals performed by any of the authors. Therefore, ethical approval was not required.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.