Bioengineering of extracellular vesicles with scaffold proteins for drug delivery

Abstract

Extracellular vesicles (EVs) play a crucial role in intercellular communication by transmitting information and participate in various physiological and pathological processes, including cancer, neurodegenerative diseases, and cardiovascular diseases. EVs have emerged as promising drug delivery vehicles, possessing unique advantages such as low immunogenicity, the ability to cross biological barriers, and a versatile cargo capacity. However, it still faces multiple challenges such as low production yield, limited targeting ability, and the complexity of using engineered EVs for clinical applications. Scaffold proteins (such as tetraspanins (TSPANs) and type I transmembrane proteins) can enhance the targeting specificity, loading capacity and stability of engineered EVs. Engineering with these proteins enables EVs to be directed to specific tissues for efficient drug delivery. Scaffold protein-based engineering of EVs has enabled the loading of various types of cargo, such as proteins, nucleic acids, and gene editors, and their targeted delivery to specific organs, such as heart and brain. It is expected to revolutionize drug delivery systems in the future and change the treatment methods for many diseases. This article reviews in detail the research progress of reported scaffold proteins and their engineered EVs by classification. We conduct a comprehensive analysis of scaffold proteins and their applications. Lastly, we discuss the current challenges and propose future directions of scaffold protein-based EV engineering.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12951-026-04142-6.

Introduction

Cell-to-cell communication is fundamental for maintaining cellular functions and tissue homeostasis in multicellular organisms. This communication can occur via direct cell-cell contact or through the transfer of secreted molecules, such as proteins, nucleic acids, and lipids, often packaged within membrane-bound structures [1]. Inspired by these natural communication pathways, drug delivery systems aim to mimic or exploit them to transport therapeutic agents with high precision to target tissues or cells, thereby minimizing off-target effects and improving therapeutic efficacy. An ideal drug delivery system not only ensures accurate targeting but also protects therapeutic cargo from degradation, maintaining its stability until it reaches the site of action.

Extracellular vesicles (EVs)—cell-derived, membrane-bound nanostructures—have emerged as highly promising candidates. Since their first description in the 1980s [2], EVs have been recognized for their ability to transport a diverse range of bioactive molecules, including DNA, RNA, proteins, lipids, and enzymes [3], from donor to recipient cells, thereby modulating cellular functions. Compared with synthetic carriers, EVs possess inherent advantages such as biocompatibility, low immunogenicity, and the ability to cross biological barriers, including the blood-brain barrier (BBB) [4]. These properties make EVs attractive platforms for targeted drug delivery, gene therapy, and regenerative medicine. Differential proteomics argues against a general role for CD9, CD81 or CD63 in the sorting of proteins into EVs.

Despite these advantages, natural EVs usually have problems such as limited targeting specificity, poor carrying capacity and poor in vivo stability, which greatly limit their clinical application. To address these challenges, various bioengineering methods have developed, aiming to enhance the performance of EVs by modifying scaffold proteins, such as key structural and functional membrane proteins like tetraspanins (TSPANs) proteins. Scaffold protein engineering technology holds great promise for improving tissue tropism, enhancing drug encapsulation efficiency and increasing delivery accuracy, making it a core strategy in the new generation of EV-based therapeutics.

In this review, we provide an overview of the biological roles and advantages of EVs as drug delivery vehicles, critically assess current engineering strategies aimed at enhancing their performance, and place particular emphasis on scaffold protein-based modifications as a versatile and powerful means of improving targeting specificity and cargo loading. We further address the key challenges that remain, outline emerging research directions, and highlight the translational potential of scaffold protein-engineered EVs for clinical applications.

Methods

This review focuses specifically on scaffold protein-based engineering of EVs, in which a defined membrane-anchored protein or protein domain is genetically fused to cargo or targeting moieties to achieve their stable display on, or loading into, EVs. For the purpose of this article, we considered as scaffold proteins classical EV-associated TSPANs (e.g., CD63, CD9, CD81), Lamp2b, transmembrane domains (TMDs) from receptors (e.g., platelet-derived growth factor receptor (PDGFR)), viral glycoproteins (e.g., vesicular-stomatitis-virus-G (VSV-G)), Gag-based viral budding scaffolds, C1C2 domains of lactadherin/milk fat globule-epidermal growth factor-factor 8 (MFG-E8), glycosylphosphatidylinositol (GPI)-anchored tags, and other proteins/domains that directly drive cargo incorporation into EVs. We excluded studies that only used chemical conjugation, lipid insertion, polymeric nanoparticles, liposome fusion, or preconditioning approaches without a genetically encoded membrane scaffold, as well as work on purely synthetic EV mimetics.

We performed a structured literature search in PubMed. We combined controlled vocabulary and free-text terms related to EVs, engineering, and individual scaffold proteins. Representative search strings included:

1. (extracellular vesicle*[Title/Abstract] OR exosome*[Title/Abstract] OR microvesicle*[Title/Abstract]) AND (engineer*[Title/Abstract] OR “surface engineering“[Title/Abstract] OR “surface modification“[Title/Abstract] OR “surface display“[Title/Abstract] OR “targeted exosome*“[Title/Abstract] OR “engineered exosome*“[Title/Abstract]) NOT Review[Publication Type].

2. ((extracellular vesicle[Title/Abstract] OR exosome[Title/Abstract] OR microvesicle[Title/Abstract]) AND (engineer*[Title/Abstract] OR modification[Title/Abstract] OR “fusion protein“[Title/Abstract]) AND ( scaffold protein[Title/Abstract] OR CD9[Title/Abstract] OR CD63[Title/Abstract] OR CD81[Title/Abstract] OR TSPAN*[Title/Abstract] OR LAMP2B[Title/Abstract] OR PDGFR[Title/Abstract] OR PTGFRN[Title/Abstract] OR plexin[Title/Abstract] OR PLXNA1[Title/Abstract] OR GPI[Title/Abstract] OR “GPI-anchored“[Title/Abstract] OR myristoylat*[Title/Abstract] OR ARRDC1[Title/Abstract] OR C1C2[Title/Abstract] OR VSV-G[Title/Abstract] OR VSVG[Title/Abstract] OR Gag[Title/Abstract] ) ) NOT Review[pt].

Because terminology in this field is highly heterogeneous, we complemented database searches with citation chasing. Several seminal studies on engineered EVs were used as “seed articles”, and their reference lists, “cited by” records, and “similar articles” suggestions were manually examined to identify additional relevant reports.

All records retrieved by these approaches were screened based on title and abstract, followed by full-text assessment for studies that potentially met the inclusion criteria. Only articles in which a defined scaffold protein was explicitly used to engineer EVs (for cargo loading, surface display, or targeting) were finally included in our mechanistic analysis and tables. Using these criteria, a total of 192 original research articles on scaffold protein-based EV engineering were ultimately included in this review. Given the diversity and inconsistency of nomenclature (e.g., many articles describe “engineered exosomes” without mentioning “scaffold”), this work should be regarded as a structured narrative review rather than a fully systematic review, and we cannot exclude that some eligible studies were not captured despite our multi-step search and manual screening strategy. A comprehensive list of scaffold protein-engineered EV studies is provided in Table S1.

Biology and therapeutic applications of EVs

EVs are heterogeneous, membrane-enclosed particles ranging from 30 to 1000 nm in diameter and are commonly classified according to their biogenesis into exosomes and microvesicles [5, 6]. Exosomes (30–150 nm) originate from the endosomal pathway and are released upon fusion of multivesicular bodies (MVBs) with the plasma membrane, whereas microvesicles (150–1000 nm) are generated by direct budding from the plasma membrane [7–9]. Both EV subtypes carry diverse bioactive cargos, including proteins, lipids, and RNAs, and play critical roles in intercellular communication [8]. EV composition is highly variable and influenced by the source cell type, physiological state, and environmental conditions, contributing to substantial EV heterogeneity [10].

EVs play a key role in intercellular communication, enabling cells to exchange information and regulate physiological functions [7]. These vesicles transport proteins, lipids, RNAs and other signaling molecules (Fig. 1), and affect recipient cells in multiple ways, including regulating their growth, differentiation and immune responses [11, 12].

Fig. 1.

EV biogenesis, composition, and uptake. Two types of EVs are released from donor cells through two primary pathways. (A) exosomes (30–150 nm), which originate from MVBs that fuse with the plasma membrane, and (B) microvesicles (150–1000 nm), which bud directly from the plasma membrane. The EV membrane harbors a wide array of proteins, including TSPANs, type I transmembrane proteins, lipid-anchored proteins, peripheral membrane proteins, and virus-related proteins, which contribute to their structural integrity and functional specialization. EVs also encapsulate diverse cargo, such as nucleic acids, metabolites, and proteins, thereby mediating intercellular communication. Recipient cells can internalize EVs through endocytosis or by receptor-ligand interactions at the plasma membrane, enabling functional transfer of EV-associated biomolecules

EVs are essential for maintaining cellular homeostasis and are involved in processes such as waste removal [13], immune modulation [14], and regulation of cellular stress [15]. In pathological contexts, EVs contribute to tumor progression, neurodegenerative disorders, and cardiovascular diseases [16]. Tumor-derived EVs can modulate immune cell function and facilitate tumor growth and metastasis [17]. While in neurodegenerative diseases, EV mediate the spread of pathogenic proteins such as tau and α-synuclein [18]. In cardiovascular diseases, EVs are involved in regulating processes such as angiogenesis, immune response, and myocardial hypertrophy [19, 20]. Collectively, these functions position EVs as central mediators of disease pathogenesis and promising platforms for therapeutic development.

EVs as drug delivery systems: advantages and challenges

Over the past decades, numerous synthetic delivery platforms—such as liposomes, lipid nanoparticles (LNPs), and micelles—have been developed to improve the pharmacokinetic and pharmacodynamic profiles of therapeutic agents [21]. However, these platforms suffer from intrinsic limitations. Liposomes show limited targeted delivery capability and preferentially accumulate in reticuloendothelial system organs such as the liver and spleen, leading to off-target deposition and immune-related adverse effects [22, 23]. In addition, LNPs are associated with polyethylene glycol (PEG)-induced accelerated blood clearance that shortens circulation time and complement activation-related pseudoallergy that compromises safety and therapeutic efficacy [24, 25].

To overcome the limitations of synthetic carriers, there has been growing interest in harnessing naturally derived delivery systems. EVs have shown significant potential as drug delivery systems due to their unique biological properties [26–28]. EVs can also cross biological barriers such as BBB and exhibit targeted delivery capabilities to specific tissues or cells [29]. The natural biocompatibility of EVs further highlights their potential as delivery carriers, making them a promising alternative for the synthesis of nanoparticles.

Low immunogenicity is one of the biggest advantages of EVs [30]. Because EVs are derived from the recipient’s own cells, they usually do not elicit a strong immune response, in contrast to synthetic nanoparticles, which are often recognized by the immune system as foreign invaders [31]. This characteristic makes EVs highly attractive in clinical applications, as immunocompatibility is a key factor that must be prioritized in the process of clinical translation [32].

Moreover, EVs also possess the ability to cross biological barriers that are typically difficult for other delivery systems to penetrate. One prime example is their ability to cross the BBB, a critical challenge in the treatment of central nervous system (CNS) diseases [33, 34]. With this characteristic, EVs represent an ideal platform for drug delivery in the brain, while the efficacy traditional delivery systems in this field is often very limited [35, 36].

Furthermore, EVs offer cargo versatility, as they can carry a wide range of therapeutic payloads. These include small molecules [37], proteins, RNA (such as siRNA and mRNA) [11, 38, 39], and even clustered regularly interspaced short palindromic repeats (CRISPR)/CRISPR-associated protein 9 (Cas9) gene-editing components [40]. This versatility makes EVs ideal for a variety of therapeutic applications, from gene therapy to protein delivery.

Another advantage lies in the inherent targeting capability of natural EVs compared to other unprocessed drug delivery particles [27]. This targeting ability primarily relies on the homing effect, which enables precise recognition and targeting through specific “navigation molecules” such as proteins, lipids, or glycans carried on their surfaces [41]. A representative example is the ability of glycans on the EV surface to bind to chemokine receptors like chemokine (C-C motif) receptor 8 (CCR8), thereby targeting cancer cells [42]. Furthermore, EVs exhibit a parental effect, meaning they naturally carry the molecular characteristics and biological functions of their source cells. When interacting with target cells, they partially reproduce or mimic the properties of the parental cell’s effects. For instance, EVs secreted by tumor cells often present tumor type-specific integrin profiles that mediate the metastasis of tumor cells to specific organs [43]. Moreover, HEK293T cell-derived EVs were found to be accumulated in subcutaneous tumors, which has been exploited to develop EV-based anticancer therapies [44]. These two effects are not independent but tightly coupled and synergistic. The parental effect provides EVs with an inherent molecular identity for targeting, while the homing effect drives the precise execution of this identity instruction. Together, they form the core logical chain of EV-mediated targeted delivery—from source design to terminal execution.

However, despite these promising advantages, several challenges remain in the use of EVs as drug delivery systems. One major challenge is their low yield [45]. The production of EVs in sufficient quantities for clinical use remains a significant hurdle, as they are typically secreted at low levels by cells. Efforts to optimize large-scale production methods are ongoing [46, 47]. Another challenge is their limited targeting ability. While EVs have inherent targeting potential, natural EVs often accumulate in non-target tissues, which can reduce their therapeutic efficacy and increase the risk of side effects [48]. The non-specificity of this distribution pattern significantly undermines the potential of EVs for precision medicine, making the enhancement of EV targeting a crucial research direction.

Furthermore, even if EVs successfully enter target cells, they remain constrained by intracellular endocytosis pathways. Research indicates that although EVs enter cells via endocytosis, they can be efficiently internalized by various cells through mechanisms such as pinocytosis and clathrin-mediated endocytosis, most of these internalized EVs are transported to the endosomal-lysosomes pathway, ultimately leading to the degradation of their therapeutic cargo [49, 50]. This mechanism significantly affects the therapeutic efficacy of the therapeutic agents loaded in these EVs [51].

To address these issues, researchers have proposed various engineering strategies. From a targeting perspective, strategies to improve EV specificity include displaying ligands or antibodies on the EV membrane [52, 53] and exploiting tissue-specific molecular recognition pathways [54]. Regarding lysosomal escape, researches have demonstrated that incorporating viral proteins such as VSV-G [55] and influenza hemagglutinin (HA) [56] into the EV membrane, inspired by viral escape mechanisms, enhances the membrane penetration capacity of endosomes [57]. The application of these strategies is expected to overcome the existing limitations of EVs, offering new insights for their use in clinical drug delivery and gene therapy.

Finally, the complexity of engineering EVs presents another challenge. Modifying EVs to carry therapeutic cargo or specific targeting peptides while maintaining their biological function is technically demanding [58]. Achieving a balance between effective cargo delivery and preserving the natural biological properties of EVs is crucial for their success as therapeutic agents [59, 60].

Therapeutic potential of EVs

EVs are increasingly being explored for diverse therapeutic applications, due to their inherent capacity to carry and deliver biomolecules [5, 61]. This makes them promising candidates for treating a wide range of diseases, including cancer, genetic disorders, and cardiovascular and neurodegenerative diseases.

One of the most significant application areas for EVs is cancer therapy. By artificially modifying EVs, they can deliver chemotherapy drugs, RNA-based therapeutics, and immunomodulatory molecules to achieve specific targeted treatment against tumors [62]. Recent studies have demonstrated that engineered EVs can carry immune checkpoint inhibitors or tumor-targeting antibodies, thereby effectively enhancing anti-cancer immunity [63]. These engineered EVs have shown the ability to target cancer cells more precisely, enhancing the therapeutic effect while reducing side effects and significantly improving treatment prognosis.

EVs also demonstrated significant potential in the field of gene therapy [64]. In particular, EV-based systems have become effective carriers for CRISPR-Cas9 gene editing tools, and can even achieve efficient genome editing in hard-to-target tissues such as the brain [65, 66]. By precisely delivering gene-editing tools, these engineered EVs offer a powerful strategy for treating genetic disorders and other conditions requiring precise targeted interventions [67].

Cardiac repair is another area where EVs have shown therapeutic potential. Studies have demonstrated that EVs derived from mesenchymal stem cells (MSCs) [68] or cardiac cells [20, 69] can effectively promote cardiac tissue regeneration and improve cardiac function following injury. These EVs contain multiple bioactive molecules that support tissue repair and cell survival, offering a promising new therapeutic strategy for treating cardiac diseases such as myocardial infarction (MI) and heart failure [30, 70].

Additionally, the potential of EVs in treating neurodegenerative diseases is also being explored. Due to their ability to cross the BBB, EVs can be used to deliver neuroprotective drugs and genetic material directly to the CNS [71]. This makes them an attractive option for treating diseases like Alzheimer’s, Parkinson’s, and Huntington’s disease, where traditional drug delivery systems fail to effectively reach the brain [21, 72]. EVs can maintain the natural biological properties of the biomolecules they carry while crossing biological barriers, opening up new prospects for personalized medicine, gene therapy, and cancer immunotherapy. As research continues to advance, EVs may become a core component of next-generation therapies, offering novel solutions for achieving more precise treatments with enhanced efficacy and reduced side effects.

Engineering of EVs

Natural EVs show great promise as novel therapeutic agents and drug delivery systems, but they still face significant limitations. Engineered EVs obtained by modifying source cells or natural EVs can serve both as direct therapeutic agents and as highly efficient drug delivery carriers [73]. Engineering of EVs is a rapidly evolving field aimed at optimizing their inherent properties to enhance the delivery capacity of therapeutic agents. Successful EV engineering requires the modification of their biological features, focusing on three primary considerations: targeting specificity, loading capacity, and biological stability. This section outlines the key factors for engineering EVs and the methods employed to enhance their therapeutic potential.

To enhance the efficiency, stability, and targeted delivery capabilities of EVs for cargo loading, various methods for engineering EVs have been developed (Fig. 2). The most widely used and common approach currently relies on cell-based genetic modification. This involves overexpressing specific nucleic acids, proteins, or scaffold proteins in donor cells, enabling their natural enrichment within secreted EVs to achieve passive loading [74]. Physical methods such as electroporation [75] and freeze-thaw [76] treatment can enhance the loading of exogenous molecules, while chemical modification [77] and hybrid vesicle technology [76] can endow EVs with new functional properties. These methods collectively form a multidimensional toolkit for EV engineering, continuously expanding its potential applications in fields such as drug delivery, gene therapy, and regenerative medicine.

Fig. 2.

Strategies for engineering EVs. EVs can be modified through multiple approaches to improve their therapeutic efficacy and targeting specificity. Passive loading via overexpression enriches EVs with proteins or nucleic acids produced in donor cells. Genetic modification of donor cells allows EVs to actively incorporate nucleic acids, proteins, scaffold proteins, or genome-editing tools. Electroporation transiently permeabilizes EV membranes to facilitate loading of nucleic acids and other macromolecules. Freeze-thaw cycles disrupt and reconstitute membranes, enabling encapsulation of diverse cargo. Chemical modification introduces functional moieties through covalent or noncovalent conjugation to EV membranes. Hybrid vesicle formation, achieved by fusing EVs with liposomes or nanoparticles, combines the advantages of natural and synthetic systems for enhanced delivery. Together, these strategies provide a versatile toolbox for tailoring EVs as platforms for drug delivery, gene therapy, and regenerative medicine

Loading capacity

The loading capacity of EVs is a key factor for their effectiveness as drug delivery systems. EVs must be engineered to carry sufficient therapeutic cargo, such as small molecules, proteins, or RNA. To enhance therapeutic drug loading efficiency while preserving the functional characteristics of EVs, several strategies are employed.

In addition to the simplest strategy of passive loading, achieved by overexpressing the desired cargo in donor cells, one common approach is to use scaffold proteins like CD63, which enhance cargo loading. For instance, when CD63 was fused with mCherry and the cargo protein was fused with an anti-mCherry nanobody, 55% of EVs are successfully loaded with the cargo [78]. These modifications significantly improve the loading capacity and delivery efficiency of EVs. Furthermore, techniques such as electroporation (increasing membrane permeability) and lipid modifications (incorporating hydrophobic molecules) are used to improve cargo encapsulation, especially for larger biomolecules such as RNA and proteins. However, these physicochemical methods may damage EVs and compromise their intrinsic properties [79–81].

Targeting specificity

Another crucial aspect of EV engineering is improving the targeting specificity of EVs to ensure they deliver therapeutic agents efficiently to specific tissues or cells, while minimizing off-target effects. The targeting ability of EVs is primarily determined by the proteins expressed on their surface, such as TSPANs [41] and integrins [82], which dictate their tropism (i.e., their ability to preferentially target certain cells or organs).

Surface modifications are the most widely adopted strategy for improving the targeting capacity of EVs. One approach is the fusion of EVs with targeting ligands, such as antibodies or peptides, allowing them to bind to specific receptors on target cells. For example, streptavidin expressed on the surface of EVs can bind to biotinylated ligands, enabling precise targeting to specific cells [83]. Another strategy involves using fusion proteins, such as nanobodies or phosphatidylserine (PS) binding proteins, which enable EVs to target tumor cells or other specific tissues, enhancing targeting precision [84]. Studies have also shown that presenting ligands on the EV surface significantly reduces tumor burden in the EV-Ligand treatment group compared to the group treated with ligand-free EVs at the same dosage [85].

Genetic engineering has become an important technique for modifying EVs. This method involves transgenic expression of specific proteins on the EV surface or within the vesicles, enhancing their functional properties. One common approach is the expression of scaffold proteins, such as Lamp2b, which can display targeting peptides on the surface of EVs. Through this mechanism, EVs can be directed to target specific tissues such as cardiac or brain cells, thereby enabling targeted drug delivery [86].

Biological stability

For engineered EVs to be clinically useful, they must remain biologically stable in the body, avoiding degradation or clearance before reaching their target sites. Biological stability ensures that EVs can effectively deliver their therapeutic cargo to the intended location [87]. Strategies to enhance stability include the use of chemical modifications and encapsulation techniques that protect the EVs from enzymatic degradation and premature clearance by the immune system [60].

A classic approach to implementing this strategy is to incorporate “don’t eat me” signals onto the EVs’ surface [88]. A prime example is the protein CD47. This transmembrane protein is naturally expressed on healthy cells and functions as an immune checkpoint. By binding to the signal regulatory protein α (SIRPα) receptor on macrophages, it transmits a “do-not-eat” signal that prevents the cell from being phagocytosed [89]. Therefore, when EVs are genetically modified to display CD47 on their surface, their circulation time in the blood is greatly prolonged, enhancing their chance of arriving at the target site [90].

Engineering EVs for therapy requires developing a range of strategies to optimize their therapeutic potential, primarily by improving their targeting specificity, loading capacity, and biological stability in biological environments. Surface modifications, genetic engineering, and cargo optimization techniques are the primary methods used to achieve these goals. By improving these key aspects, EVs can become more effective delivery vehicles for a wide range of therapeutic applications, including cancer treatment, gene therapy, and tissue regeneration [91]. As research in EV engineering advances, these techniques hold the potential to revolutionize drug delivery systems and pave the way for more effective and targeted clinical therapies [92].

Scaffold proteins for EV engineering

To further optimize the therapeutic potential of EVs, the engineering based on scaffold proteins plays a critical role in enhancing their targeting specificity, loading capacity, and biological stability [93]. As shown in Fig. 3, scaffold proteins are key components that can significantly improve the functionality of EVs, making them more effective delivery vehicles for therapeutic applications [94]. These proteins, through surface modifications or genetic engineering, enable EVs to target specific cells or tissues more precisely, while also improving the efficiency with which they carry and release therapeutic cargo [58].

Fig. 3.

Scaffold protein-based engineering of EVs for enhanced cargo loading and targeting specificity. EV engineering strategies exploit diverse classes of scaffold proteins to improve loading efficiency and targeting precision. (a) TSPANs (e.g., CD9, CD63, CD81) can be fused with tags or RNA/protein-binding domains to facilitate selective cargo loading and targeting of recipient cells or viruses. (b) Type I transmembrane proteins (e.g., prostaglandin F2 receptor inhibitor (PTGFRN), Lamp2b, PDGFR) provide dual anchoring and targeting functions, coupling extracellular ligand- or antibody-binding domains with intracellular cargo-loading motifs. (c) Lipid-anchored membrane proteins (e.g., GPI, N-myristoylated motifs) insert into the EV bilayer to tether functional molecules or cargos such as Cas9. (d) Peripheral membrane proteins (e.g., arrestin domain-containing protein 1 (ARRDC1), C1C2 domain of MFG-E8) associate with the EV surface via electrostatic or domain-specific interactions, acting as modular adaptors for indirect cargo recruitment or surface functionalization. (e) Virus-related proteins (e.g., Gag, VSV-G) contribute fusogenic and tropism-enhancing properties, promoting EV uptake, endosomal escape, and improved targeting, and they can also function as scaffolds for efficient cargo encapsulation. Collectively, these strategies demonstrate how scaffold proteins enable rational EV engineering to advance drug delivery, gene therapy, and regenerative medicine

By modifying the surface of EVs with scaffold proteins, researchers can direct these vesicles to specific sites of action, such as tumors or damaged tissues, ensuring that drugs or genetic materials are delivered more efficiently [95]. Additionally, scaffold proteins can enhance the cargo-loading capacity of EVs, enabling them to carry a diverse range of therapeutic agents, including small molecules [96], proteins [97], and RNA [93, 98]. Moreover, scaffold proteins also contribute to the stability of EVs in the body, preventing degradation or clearance, which is crucial for effective drug delivery [81].

As research progresses, the role of scaffold proteins in EV engineering is becoming increasingly important, offering exciting opportunities for improving drug delivery systems and advancing clinical therapies in areas like cancer treatment (e.g. Codiak’s exo^IL‑12 EV using PTGFRN scaffold, entering Phase I trial) [99], gene therapy (e.g. KRAS‑targeting siRNA‑loaded MSC‑EVs in Phase I pancreatic cancer trial) [100], and tissue regeneration [101]. The continued exploration of scaffold proteins in EV engineering holds great promise for revolutionizing the way we approach targeted drug delivery and precision medicine.

Loading

TSPANs

TSPANs are found in almost all tissues and cell types, with each member exhibiting different expression profiles. For example, TSPANs CD9, CD63, CD82, and CD151 are widely distributed across various cell types, while CD37 and CD53 are primarily found in leukocytes [102]. TSPANs, such as CD9, play important roles in the immune system. They interact with various key leukocyte proteins, including immune receptors, integrins, and signaling molecules, and may play a crucial role in tumor immune surveillance [103]. CD9, CD81, and CD63 are major components of EVs, and studies have found that CD9, CD81, and CD63 play a role in determining the protein composition of EVs released from MCF7 breast cancer cells, emphasizing their regulation of the expression and transport of immunoglobulin (Ig) domain proteins such as CD9P-1 and EWI-2 [104].

CD63

CD63 is a widely expressed member of the tetraspanin family that contributes to the formation of lipid-rich microdomains containing gangliosides and cholesterol, thereby organizing networks of partner proteins and regulating their transport, cleavage, and functional interactions [105]. Unlike most other TSPANs, CD63 is predominantly localized to late endosomes, lysosomes, and microvascular endothelial regions [106], and it is especially enriched in EVs, which has established CD63 as a reliable marker and a practical tool for manipulating this vesicle subpopulation [107]. Beyond its role as a marker, CD63 participates in diverse cellular processes including cargo sorting and intracellular trafficking. For example, it regulates C-X-C chemokine receptor type 4 targeting to lysosomes in T cells [105]. Structurally, TSPANs have a conserved “M” topological configuration, which consists of two short cytoplasmic termini, one small intracellular loop, two extracellular loops, and four TMDs [108]. This organization provides multiple engineering opportunities for EV design. Using a fluorescence labeling strategy as an example, researchers constructed a CD63-GFP-CD63-RFP fusion construct with RFP fused to the extracellular loop, while GFP was fused to the intracellular loop yielded fluorescence and co-localized signals in HEK293 cells, confirming the successful presentation of fluorescent markers on both surfaces of EVs [109]. Further studies revealed that the large extracellular loop (LEL) of CD63 plays a crucial role in its recruitment to EVs. When the LEL of CD63 fused with mCherry was deleted, the protein was no longer found in the purified EV fraction [110]. In addition, transmembrane helix 3 (TM3) of CD63 is essential for EV anchoring and targeting. Systematic deletion of the other TMDs generated a series of CD63 truncates that adopt distinct topologies in the EV membrane. Notably, provided TM3 is preserved, these truncates maintain potent membrane anchoring and EV-targeting activity, and TM3 alone is sufficient for targeting, providing a minimal scaffold [111].

Fusion expression approaches have been widely adopted to exploit CD63 as a scaffold, with protein cargos serving as a representative example [112]. For example, by fusing Cre recombinase with CD63, efficient loading into the lumen of EVs can be achieved. In parallel, CD63 can support active mRNA loading through the incorporation of high-affinity RNA-binding domains. In addition, alternative affinity-based systems can also be employed to enhance cargo loading, such as the RNA aptamer-aptamer-binding protein (ABP) and protein-antibody interaction-based strategies.

Using GFP as a reporter cargo, multiple comparative studies have shown that CD63 exhibits superior loading performance, as evidenced by the enhanced EV loading capacity, with GFP-CD63 expression levels approximately 6-fold higher than those of untagged GFP in EVs [113, 114]. Different scaffold proteins exhibit pronounced subpopulation specificity in EV loading, with CD63-GFP showing the highest enrichment efficiency in CD63⁺ EVs (59% ± 2%) [115]. Extensions of this concept include luciferase fusions, which allowed the generation of sensitive bioluminescent labeling systems for real-time EV tracking in vitro and in vivo [116, 117]. For example, the CD63- ThermoLuc (Tluc) fusion provides a highly sensitive, quantitative readout for EV release, essential for high-throughput screening [118]. Moreover, the co-expression of CD63-Intein-Cre fusion proteins with viral fusion proteins markedly enhanced Cre delivery, achieving GFP expression levels of 66% and 98% in HeLa-TL and T47D-TL cells, respectively, after two days of EV treatment [65]. For RNA loading, the MS2-MCP system can be employed to recruit sgRNA into EVs, thereby facilitating the enrichment of CRISPR-Cas9 ribonucleoprotein (RNP) [119]. Moreover, fusing CD63 with high-affinity RNA-binding domains such as PUFe or L7Ae enables efficient enrichment of target mRNA into EVs [120–123]. In addition, selective RNA enrichment into EVs can be achieved by fusing CD63 with an RNA-binding protein (HuR), and introducing HuR-recognizable sequences (AU-rich elements, AREs) into the target RNA, with the relative expression level of artificial circular non-coding RNAs (acircRNAs) reaching up to 1000 [124]. For loading genome-editing cargos, fusion of GFP nanobodies with CD63 facilitated the selective recruitment of Cas9 into EVs [125]. Similarly, the RNA aptamer com is fused to sgRNA, while the ABP Com is fused to the EV scaffold protein CD63. Through the specific com-Com interaction, CRISPR RNPs can be efficiently enriched within EVs, resulting in approximately a five-fold enrichment of Cas9 [126]. More recently, an advanced design fused mCherry to the luminal domain of CD63 while displaying a 3×FLAG tag externally, allowing specific recruitment of mCherry nanobody-fused cargos. The proportion of EGFP-positive vesicles obtained using the anti-mCherry nanobody was over 13-fold higher than that obtained using a non-specific nanobody. Consistently, EVs loaded with Cre via an anti-mCherry nanobody generated a higher proportion of iRFP-positive cells, indicating enhanced loading and delivery of Cre [78].

Overall, CD63 stands out as one of the most versatile and extensively validated scaffolds for EV cargo loading. Its strong endogenous enrichment in late endosome-derived EVs, together with well-defined luminal topology, makes CD63 particularly suitable for applications requiring high-efficiency intraluminal loading, including reporter proteins, recombinases, Cas9 RNPs, and actively enriched RNAs. Consequently, CD63-based designs are especially well suited for mechanistic studies, quantitative loading comparisons, and proof-of-concept demonstrations of EV-mediated functional delivery.

CD9

CD9, also known as Tspan 29, is a tetraspanin with a molecular weight of 21–24 kDa [127] that has been extensively exploited as a scaffold protein for engineering EVs. Its versatility is illustrated by applications ranging from nucleic acid and protein enrichment to genome-editing delivery.

CD9 enables EV cargo loading either by directly binding cargo molecules or indirectly by fusing with functional protein modules that recruit cargo. For example, GFP can be fused to CD9 via genetic fusion, and the resulting CD9-GFP fusion protein is commonly used for EV labeling and tracking [128]. Similarly, by fusion expression with CD9, RFP can be efficiently sorted and loaded into EVs, and subsequently delivered to recipient cells via EV-mediated transfer [129]. In addition, diverse photo-responsive systems can be incorporated to enable CD9-mediated, controllable cargo recruitment and loading. The EVs for protein loading via optically reversible protein-protein interactions (EXPLORs) system combined CIBN-EGFP-CD9 with cargo protein-cryptochrome 2 (CRY2), achieving highly efficient protein delivery, with more than 95% of recipient cells expressing EGFP following Cre-EGFP: EXPLORs treatment [130]. Similarly, CD9 has been shown to recruit Cas9-CRY2, with the CIBN-CD9 platform enabling the incorporation of up to 20 Cas9 molecules per EV, representing one of the most efficient systems for CRISPR-Cas9 loading to date [131]. More recently, mMaple3-mediated protein tethering via CD9 enabled efficient encapsulation and release of dCas9 RNP complexes targeting β-site amyloid precursor protein-cleaving enzyme 1 (BACE1), resulting in robust suppression of BACE1 and mitigation of Alzheimer’s disease-like pathology [132]. Moreover, by constructing a CD9-photocleavable protein (PhoCl)-cargo fusion structure, soluble proteins can be actively loaded into EVs and subsequently released from the CD9-anchored structure upon 405 nm light irradiation, enabling controlled protein release and delivery [133]. For RNA enrichment, the MS2-MS2 bacteriophage coat protein (MCP) system is a commonly used and well-established strategy. Through the interaction between the CD9-MCP fusion protein and the low-density lipoprotein receptor (Ldlr) aptamer containing MS2 stem-loop domains, selective enrichment of Ldlr mRNA in engineered EVs is achieved [134]. The combination of the MS2-MCP loading strategy with PhoCl-mediated light-triggered release markedly enhanced EV-mediated functional delivery of Cas9, increasing the recombination efficiency in reporter cells from approximately 2% to approximately 28% [135]. Fusion of CD9 with the RNA-binding proteins HuR or AGO2 further highlights its potential for RNA enrichment, enabling active recruitment of RNA into EVs and efficient transfer to recipient cells [136, 137].

CD9 represents a highly engineerable scaffold capable of supporting efficient protein and CRISPR-Cas cargo loading, particularly when combined with inducible dimerization or light-responsive systems. However, a key limitation of CD9 lies in its pronounced cell type-dependent EV sorting behavior, which can substantially constrain its generalizability across producer cell lines. For example, CD9 fusion constructs can be efficiently incorporated into EVs in HEK293 cells, whereas their incorporation into EVs is markedly reduced or undetectable in tumor cell lines such as HeLa and 5637 [129].

CD81

CD81 has been investigated as a scaffold protein to enhance the loading capacity of EVs. Engineering of CD81 is commonly achieved through genetic fusion strategies to facilitate cargo loading in EVs. Similar to CD63, CD81 can be fused to GFP through genetic fusion strategies, resulting in approximately 15–25% of EVs being labeled with GFP [116]. Moreover, CD81 has been reported to be highly enriched in EVs and has been used as a scaffold protein to mediate the incorporation of mNeonGreen into EVs [138]. In addition to direct genetic fusion, CD81 can also mediate cargo loading through affinity-based systems. For example, under treatment with 100 nM rapamycin, FK506-binding protein (FKBP12)-rapamycin binding (FRB)-Cre-HiBiT exhibited approximately 70-fold enrichment in EVs. Furthermore, CD81-EV-mediated delivery of Cre can be achieved through FRB-FKBP interaction, leading to Cre-loxP recombination and activation of reporter gene expression in recipient cells [139].

CD81 has been less frequently exploited for bulk cargo loading than CD63 or CD9; however, several studies have demonstrated that CD81 can support inducible or affinity-based recruitment of protein cargos under specific experimental designs. These features suggest that CD81 can be engineered to support controlled or inducible cargo recruitment, rather than serving as a scaffold that has been demonstrated to achieve maximal cargo density.

Other TSPANs

TSPANs represent an important class of scaffold proteins with diverse roles in EV biology. For example, by genetically fusing GFP to TSPAN14, a high level of cargo loading into EVs was achieved, with an average of approximately 173 GFP molecules per EV [140]. The endogenous EV loading capacity of different scaffold proteins was systematically compared and screened by genetically fusing Tluc to candidate scaffolds. The results showed that EVs engineered with TSPAN2, TSPAN3, as well as the commonly used scaffold protein CD63, exhibited higher levels of Tluc, indicating their superior protein loading capacity [93]. Subsequently, TSPAN2 and TSPAN3 were found to exhibit strong enrichment into EVs in the process of validating the sorting capacity. Engineering of TSPAN2 enabled efficient delivery of Cre recombinase in melanoma xenograft models, leading to 30% recombination in tumor cells, thereby underscoring its utility for therapeutic EV design [141].

Beyond canonical TSPANs, several less extensively studied members of the tetraspanin family have exhibited relatively high EV enrichment in specific studies. However, their application remains limited, and their compatibility with genome-editing cargos or inducible dimerization systems (such as FRB-FKBP-based recruitment strategies) has not yet been systematically evaluated. Consequently, these TSPANs are currently regarded as underexplored EV scaffolds with considerable potential for further development in engineered loading strategies and EV-based therapeutic applications.

Type I transmembrane protein

Single-pass transmembrane proteins are membrane proteins that span the lipid bilayer only once. Based on the orientation of their N- and C-termini across the membrane, they can be further classified into Type I, II, III, and IV [142]. Type I transmembrane proteins are the most common and characterized by an extracellular, or luminal, N-terminus followed by a single transmembrane helix and a cytosolic C-terminus [143]. Type I transmembrane proteins in cells can serve to transduce signals and modulate cellular behavior, thereby regulating key biological processes such as cell adhesion, migration, proliferation, and differentiation [144].

Lamp2b

Lamp2 (lysosome-associated membrane protein 2) is an important component of the lysosomal membrane, constituting approximately half of the lysosomal membrane protein pool together with Lamp1, and deficiency of Lamp2 leads to more severe cellular consequences than the loss of Lamp1 [145]. The Lamp2b isoform has been extensively exploited for EV engineering due to its strong membrane localization and ability to present functional moieties on the vesicle surface.

Cargo loading strategies based on Lamp2b are mainly achieved through fusion expression, including direct fusion of the cargo to the scaffold protein, or indirect cargo loading by linking binding domains to the scaffold. For protein cargo loading, the most commonly used strategy is direct fusion of the cargo to scaffold proteins [146]. For example, genetic fusion of GFP to Lamp2b enables GFP to be incorporated into EVs, which can be used to visualize the uptake of EVs by target cells [147, 148]. Moreover, by genetic fusion to Lamp2b, irisin was successfully loaded into EVs. The engineered EV-irisin exhibited enhanced heat tolerance, including a delayed onset of heat syncope (109 ± 8 min vs. 101 ± 8 min in the irisin polypeptide group), faster recovery time (16 ± 6 min vs. 25 ± 7 min in the irisin polypeptide group), and improved survival rates, indicating that EV-mediated delivery of irisin provides better protection against exertional heat stroke (EHS) [149]. For RNA delivery, the most widely used strategy relies on indirect fusion, in which RNA-binding domains are fused to scaffold proteins to mediate cargo loading. Early work demonstrated that adding an HA tag to MS2 and then linking MS-HA dimers to Lamp2b, resulted in significant enrichment of target RNA molecules within EVs [150]. In addition, siRNA loading can be enhanced by conjugation of an RNA recognition motif to Lamp2b, which facilitates siRNA enrichment in EVs and results in effective downregulation of GPX4 and DHODH protein expression [151]. Moreover, fusion of HuR to Lamp2b enabled the recruitment of miR-155 and its trafficking to the lysosomal pathway, resulting in a significant reduction of miR-155 expression (approximately 2 vs. 4 compared with controls) and marked attenuation of liver fibrosis [152].

Overall, Lamp2b is a multifunctional scaffold protein that can be engineered to enable cargo loading through direct fusion or indirect recruitment strategies, including the incorporation of protein- or RNA-binding modules. However, existing studies have predominantly focused on using Lamp2b for surface display of specific molecules to enhance EV targeting. Its potential to integrate both cargo loading and targeting functions remains to be further explored from an engineering perspective.

PDGFR

PDGFR is a single-chain transmembrane glycoprotein belonging to the type III tyrosine kinase receptor family, with two major subtypes, PDGFRα and PDGFRβ [153, 154]. Its structural properties make it a suitable candidate for use as a scaffold protein in EV engineering.

The PDGFR can also facilitate efficient EV sorting. For instance, fusion of CD63 and CD81 to the PDGFRβ TMD markedly enhanced sorting efficiency, with CD81 TMD-containing EVs showing an average of approximately 103 GFP molecules per vesicle [140].

Similar to Lamp2b, current studies on PDGFR have largely focused on its targeting capability, whereas its potential as a scaffold for EV cargo loading remains underexplored. The limited number of studies focusing on PDGFR-mediated cargo loading indicates that this protein has primarily been investigated in the context of EV sorting, leaving its loading potential largely unexplored.

Plexin

Plexins, first described in 1995 [155], are a family of transmembrane receptors comprising four subfamilies in vertebrates and nine members in total, including plexin-A1(PLXNA1) to A4, B1 to B3, C1, and D [156]. Recent studies have highlighted their potential as scaffold proteins for EV engineering. A truncated form of PLXNA1 has been shown to possess superior EV-sorting capacity and to permit the fusion expression of proteins of interest (POIs), thereby providing a versatile platform for molecular incorporation into EVs. Building on this property, a universal EV system was established by fusing truncated PLXNA1 with the nanobody of ALFA tag named NbALFA. This design leverages the high affinity and selectivity of the ALFA tag/NbALFA pair, enabling NbALFA to be displayed on the EV surface via PLXNA1 (residues 863–1316). The engineered EVs can subsequently be incubated with ALFA-tagged POIs, allowing for straightforward and efficient labeling [157]. This ALFA tag/NbALFA-based system illustrates the potential of plexin-derived scaffolds to generate broadly applicable engineered EVs, with considerable promise as carriers in therapeutic delivery and drug development.

Plexins, particularly truncated PLXNA1, have emerged as novel EV scaffold proteins and have demonstrated high cargo-loading capacity. In addition to conventional direct fusion strategies, this system introduces modular cargo-loading approaches, providing evidence from multiple angles to support its loading capability. Moreover, for other scaffold proteins, systematic investigations into the EV sorting mechanisms and cargo-loading capacities of truncated scaffold variants remain an important direction that warrants further exploration in scaffold protein engineering.

PTGFRN

PTGFRN protein plays an important role in regulating the activity of the PTGFR. PTGFRN is associated with cell proliferation, tumor growth, radiation resistance, and DNA damage repair [158], and contributes to maintaining physiological balance under both normal and pathological conditions. In recent years, PTGFRN has gained recognition as an effective scaffold protein for engineering EVs. Evidence supporting this role includes an investigation of candidate scaffold proteins in which GFP was fused to PTGFRN. Flow cytometry analysis revealed that nearly 95% of the particles expressed detectable GFP, underscoring the strong cargo-loading capacity conferred by PTGFRN [159]. Beyond that, truncated PTGFRN variant (PTGFRN-Δ687) has been used as an EV scaffold to enable mRNA delivery through engineered RNA-binding modules [160]. Further study has incorporated PTGFRN-Δ687 into advanced EV functionalization strategies to enhance cargo loading efficiency. By fusing the human Fc region with the intracellular region of PTGFRN-Δ687 and anchoring it on the EV membrane surface to construct an Fc/Spa interaction system can promote SpCas9 protein packaging approximately 4 times that of natural packaging (41 ng vs. 10 ng) [161].

PTTG1IP

The pituitary tumor-transforming 1 interacting protein (PTTG1IP, also known as PTTG1-binding factor, PBF) was initially identified as a 22-kDa oncogene that binds to securing [162]. As a small, single-pass N-glycosylated transmembrane protein, PTTG1IP has recently emerged as a promising scaffold for EV engineering. Its stable localization within EV membranes is critically dependent on N-glycosylation at two arginine residues, enabling efficient anchoring and loading of therapeutic cargos. Leveraging this property, PTTG1IP-based engineering approaches have been shown to mediate the functional delivery of proteins such as Cre recombinase into recipient cells and xenograft tumor models, as well as the efficient transfer of Cas9-sgRNA complexes into target cells. Correspondingly, the results showed that Cre-mediated recombination efficiencies ranged from 46% to 90%, while genome editing of the endogenous RNF2 target by Cas9-sgRNA complexes reached efficiencies of up to 6% [163].

Collectively, these findings highlight the considerable potential of PTTG1IP as a scaffold protein for advancing EV-based strategies in gene editing and therapeutic protein delivery.

Lipid-anchored membrane proteins

Lipid-anchored membrane proteins are a class of proteins that associate with membranes through covalent lipid modifications rather than transmembrane helices. These proteins are typically localized to one side of the membrane without spanning the lipid bilayer, and their membrane association primarily relies on the insertion of lipid modification motifs into the lipid bilayer, thereby enabling membrane localization and loading enrichment of cargo molecules. Due to the high modularity and generalizability of this strategy, it can accommodate a wide range of protein cargos. For example, fusion of CIBN to a CaaX motif allows controllable recruitment and EV loading of proteins through the CIBN-CRY2 dimerization system [164].

N-myristoylation

N-myristoylation (N-Myr) is an irreversible co- or post-translational lipid modification in which a myristoyl group is covalently attached to the N-terminal glycine of a substrate protein [165]. This modification functions as a membrane anchor, thereby regulating protein localization, intracellular trafficking, and biological activity. Recently, this mechanism has been adapted to promote the encapsulation of proteins into EVs, expanding the repertoire of engineering strategies for EV-based drug delivery [81, 166].

One of the earliest demonstrations of this approach involved fusion of an Src-derived N-terminal octapeptide to Cas9, enabling its myristoylation and subsequent recruitment into EV membranes. When co-expressed with the fusogenic glycoprotein VSV-G, these engineered vesicles efficiently packaged Cas9/sgRNA complexes and mediated robust genome editing in recipient cells, achieving up to 42% eGFP knockout with minimal off-target activity [167], or enabling effective silencing of the androgen receptor (AR) gene in prostate cancer (PCa) cells [168]. Building upon this principle, Ilahibaks and colleagues designed a heterodimerization-based strategy termed TOP-EVs. In this system, FKBP12 was membrane-anchored via N-Myr and paired with a T82L mutant FRB domain fused to cargo proteins through a flexible linker. The FKBP12-FRB interaction, combined with VSV-G mediated membrane fusion, facilitated efficient incorporation of diverse protein cargos-including GFP, Cre recombinase, and Cas9-into EVs, achieving Cre-mediated recombination efficiencies of up to 96 ± 2% in T47D cells and Cas9-mediated genome editing efficiencies of 29 ± 3% [169]. Subsequent application of the same design enabled anchoring of Cas9 to EV membranes, resulting in approximately 28% genome editing efficiency in Cas9 stoplight reporter cells. Systemic administration in mice further achieved effective inactivation of the Pcsk gene in hepatocytes, accompanied by reduced Pcsk mRNA levels, upregulation of LDL receptors, and enhanced LDL-C uptake [170].

A complementary strategy has exploited light-inducible heterodimerization to drive Cas9 encapsulation. The engineered EVs efficiently delivered Cas9/sgRNA complexes and mediated functional genome editing in a hepatitis B virus (HBV) replication mouse model, leading to reduced viral antigen expression and decreased levels of covalently closed circular DNA [171].

Collectively, these studies establish N-Myr as a powerful and versatile strategy for enhancing membrane targeting and EV encapsulation of therapeutic proteins. In particular, ligand- or light-induced dimerization further augments the efficiency and selectivity of cargo recruitment, broadening the engineering toolkit available for the design of EV-based delivery systems.

BASP1

The brain acid-soluble protein 1 (BASP1) is a 23 kDa acidic protein originally isolated from rat and chicken brain. It is predominantly expressed at the axonal terminals of neurons, where it participates in synaptic growth, maturation of the actin cytoskeleton, and organization of plasma-membrane architecture. N-Myr enables BASP1 to associate with the cell membrane [172]. BASP1 also demonstrates significant utility as a scaffold protein for engineering EVs. By employing BASP1 as a scaffold protein to deliver minimal RNA-induced silencing complex (minRISC) via EVs, efficient gene silencing was achieved. Compared with full-length BASP1, BASP-N10 (the first 10 residues from BASP1)-mediated minRISC-EVs exhibited markedly enhanced EGFP silencing, with a knockout efficiency reaching 95%. In addition, knockdown of inducible nitric oxide synthase by minRISC-EVs significantly reduced pro-inflammatory cytokine production, demonstrating the anticipated anti-inflammatory regulatory effect. This approach provides strong support for EV-based RNA interference therapies [173].

MARCKS protein family

The myristoylated alanine-rich C kinase substrate (MARCKS) family of proteins, including MARCKS and MARCKS-like protein 1 (MARCKSL1), is a group of membrane-associated, myristoylated proteins that play crucial roles in regulating cell motility, cytoskeletal dynamics, and signal transduction [174].The MARCKS protein family is abundant in EVs derived from multiple cell sources, making it a prime candidate for further scaffold studies. Research has demonstrated that overexpressing MARCKS family proteins significantly enhances the ability of exogenous proteins and nucleic acids to enter EVs compared to traditional scaffolds, enabling high-density loading of diverse macromolecules into EVs [159].

Peripheral membrane proteins

ARRDC1

ARRDC1 is a membrane-associated adaptor that recruits the endosomal sorting complex required for transport-I (ESCRT-I) subunit tumor susceptibility gene 101 (TSG101) to the plasma membrane, thereby initiating the release of EVs enriched in TSG101, ARRDC1, and other cellular proteins [175]. Fluorescent labeling studies have shown that ARRDC1 is predominantly localized to the plasma membrane, and its overexpression promotes the redistribution of TSG101 to the cell surface, highlighting its role in vesicle biogenesis [176].

ARRDC1 has been investigated as a scaffold for loading therapeutic proteins into EVs. One notable example is the engineering of ARRDC1 to deliver p53. By constructing a fusion vector in which ARRDC1 was directly linked to the N-terminus of wild-type p53, researchers successfully incorporated the tumor suppressor into ARMMs (ARRDC1-mediated EVs). In p53-deficient cells, expression of the ARRDC1-p53 fusion markedly enhanced the transcription of canonical p53 target genes such as MDM2 and p21, demonstrating that the loaded protein retained its biological activity and could restore p53-dependent signaling [177]. Using ARRDC1 as a scaffold protein, functional cargos were efficiently packaged into ARMMs and further engineered for cell-specific delivery by surface display of Nipah virus G and F proteins coupled to targeting ligands. This engineered platform achieved approximately 60% and 40% A-to-G base editing in CD8⁺ T cells (vs. approximately 10% in CD8⁻ cells), and Cre-mediated recombination in over 50% of PV⁺ neurons [178]. Moreover, ARMMs have considerable promise as a novel platform for packaging and delivering CRISPR-Cas9 complexes. By fusing Cas9 to ARRDC1, efficient Cas9 encapsulation can be achieved, and ARMM-mediated gene editing efficiency was concentration-dependent, reaching up to 90% [179]. Although ARRDC1 has been explored as a scaffold for RNP loading and delivery, one study has reported that the effect of ARRDC1 is not as good as that of the tPA secretion peptide., indicating that further investigation and optimization are required [180].

These findings establish ARRDC1 as not only a key regulator of vesicle budding but also a versatile scaffold for engineering ARMMs to deliver functional proteins. Its ability to recruit ESCRT machinery and package bioactive molecules underscores its potential utility in therapeutic EV design.

C1C2

In addition to commonly used scaffold proteins, the PS-binding domain (C1C2) domain has also been widely applied in EV engineering. Derived from the N-terminal fragment of MFG-E8, the C1C2 domain binds to PS, thereby anchoring fused molecules onto the EV membrane [181]. The C1C2 domain can function as a scaffold protein that mediates cargo loading into EVs. By exploiting the membrane-anchoring property of the C1C2 domain, reporter proteins such as mCherry and Luciferase can be fused to C1C2 and passively incorporated into EVs during vesicle biogenesis [182, 183].

Virus related proteins

Virus-derived proteins represent a unique class of scaffolds for EV engineering, owing to their intrinsic ability to mediate membrane remodeling, cargo encapsulation, and efficient entry into host cells [115]. Viruses have evolved specialized structural and envelope proteins that enable them to overcome biological barriers, such as membrane fusion, intracellular trafficking, and endosomal escape—challenges that closely parallel those faced by engineered EVs as drug delivery vehicles [65]. Compared with native EV proteins such as TSPANs or integrins, virus-related scaffolds often endow EVs with superior fusogenicity, membrane penetration, and tissue tropism, making them particularly valuable for applications requiring efficient cytoplasmic delivery of macromolecular therapeutics (e.g., mRNA, siRNA, CRISPR-Cas9) [115]. However, their use also introduces potential concerns, including immunogenicity, safety, and scalability, which must be carefully addressed for clinical translation [35]. VSV-G in particular elicits robust B-cell humoral responses, generating neutralizing antibodies even after a single dose and thereby limiting the feasibility of repeat administration, especially in individuals with prior exposure [184]. While these issues pose significant challenges for chronic or multi-dose therapeutic regimens, virus-based scaffolds may remain compatible with one-shot applications such as in vivo genome editing. For example, engineered DNA-free virus-like particles (eVLPs) have achieved robust therapeutic editing following a single injection [185]. To reduce the risk of virus-derived scaffolds such as VSV-G being recognized and neutralized by host antibodies, several engineering strategies can be employed. For example, introducing or enhancing glycosylation on exposed regions of the viral envelope protein can create a “glycan shield” that masks immunodominant epitopes, thereby decreasing antibody accessibility and delaying immune clearance [186]. In addition, the use of heterologous vesiculovirus G proteins can circumvent humoral immunity during in vivo gene delivery [187].

VSV-G and other viral glycoproteins

The functional loading capability of viral glycoproteins, when used as scaffold proteins, for protein and other molecular cargos has been validated in multiple studies [188]. VSV-G is widely employed in the design of engineered EVs and virus-like particles due to its ability to promote membrane fusion and endocytosis. This fusogenic capacity enhances vesicle uptake by recipient cells and facilitates the cytoplasmic release of encapsulated macromolecules, thereby improving the delivery efficiency of proteins and nucleic acids. Experimental evidence has shown that VSV-G not only enhances the functional uptake of cargo but can also increase vesicle yield. For example, co-expression of cyclic GMP-AMP synthase (cGAS) with VSV-G was reported to improve both the production of EVs and the incorporation of cGAS protein into these vesicles, demonstrating that VSV-G can act as a dual enhancer of EV biogenesis and cargo packaging [55].

In addition to its classic function of promoting membrane fusion, VSV-G has also been found to serve as an engineered scaffold for enhancing the functional characteristics of EVs. In one study, an aptamer-binding protein (Com) was fused to the C-terminus of VSV-G to construct the VSV-G-Com construct. Although this approach enabled aptamer-based cargo interactions, the gene editing efficiency of the obtained vesicles decreased. This might due to the reduced expression of VSV-G or interference with its membrane fusion activity, which in turn may have limited endoplasmic reticulum escape in recipient cells [126]. VSV-G has also been engineered to load GFP into EVs as a reporter cargo, enabling visualization, tracking, and affinity capture [189], while markedly increasing the proportion of GFP-positive recipient cells compared with traditional EVs, thereby highlighting the capacity of VSV-G in enhancing cargo delivery and cell uptake [115].

More sophisticated engineering has sought to directly link VSV-G to therapeutic proteins using protein-splicing systems. One notable example is the VFIC platform (VSV-G-Foldon-Intein-Cargo), in which VSV-G is fused to Cre recombinase or Cas9 via a mini-intein. This design yielded EVs with markedly improved loading and delivery capacity. In vitro experiments showed that VFIC EVs mediated Cre recombinase transfer with nearly complete GFP reporter activation and achieved genome-editing efficiencies of approximately 80% when delivering Cas9/sgRNA complexes, far surpassing the performance of conventional EVs [65]. In addition, fusion of glucocerebrosidase (GBA) to VSV-G resulted in a significant increase in intracellular GBA activity, with approximately 45% and 40% increases observed for GBA-VSV-G-GFP and VSV-G-GFP-GBA, respectively, compared with unmodified controls [190]. These findings emphasize the versatility of VSV-G as both a fusogenic enhancer and a functional scaffold in the context of EV-mediated cargo delivery.

Similarly, the viral glycoprotein CNV-G, derived from a viral source distinct from VSV-G, has also been employed for cargo loading and delivery in engineered EVs. CNV-G modification enables efficient encapsulation of functional Cre recombinase and its subsequent delivery to neurons, resulting in robust genome recombination. Compared with VSV-G-modified EVs, CNV-G-modified EVs exhibit greater accumulation in the brain and spleen [191]. Together, these findings indicate that viral glycoproteins can serve as fusogenic transmembrane scaffolds with the capacity to support functional cargo loading while enhancing EV-mediated delivery performance.

Gag

The Gag protein is the major structural component of retroviruses such as HIV-1, where it governs viral assembly, budding, and maturation. Following viral budding, the Gag precursor undergoes proteolytic cleavage into matrix (MA), capsid, nucleocapsid, and several smaller polypeptides [192]. The MA domain is particularly important, as its N-Myr and basic residues enable Gag to associate with the plasma membrane and drive multimerization [193]. This intrinsic capacity for membrane binding and self-assembly not only underlies its essential role in viral particle formation but also provides a useful scaffold for EV engineering [194]. By harnessing these properties, Gag can facilitate the encapsulation of exogenous macromolecules into EVs or virus-like particles [195], such as protein [196] and RNP [197], thereby expanding the toolkit for therapeutic delivery systems. For example, gLuc activity of Gag-gLuc EVs was 133-fold higher than that of CD63-gLuc EVs [198]. Moreover, Gag-mediated RNP delivery has been extensively validated across multiple studies [199, 200]. By incorporating Gag protein for further modification of the prime editor engineered virus-like particle (PE-eVLP) construct, v3 and v3b PE-eVLPs were generated, resulting in a 65- to 170-fold increase in editing efficiency in human cells. A single injection of v3 PE-eVLPs achieved precise editing in the retina, restored protein expression, and partially recovered visual function [201].

One prominent application of Gag-based scaffolds is the nanomembrane-derived EVs for the delivery of macromolecular cargo (NanoMEDIC) platform. This system employs a chemically inducible dimerization strategy to regulate the encapsulation of SpCas9. Gag is fused to FKBP12 while SpCas9 is fused to FRB; upon exposure to the small molecule AP21967, FKBP12-FRB heterodimerization occurs, promoting efficient loading of SpCas9 into EVs [66]. The resulting vesicles not only package and transport the nuclease effectively but also enhance its release and nuclear translocation in recipient cells, enabling precise genome editing. Meanwhile, using the NanoMEDIC system for CRISPR-Cas9 RNP loading and delivery achieved 74% indel formation in HEK293T cells and induced 7 ± 1% exon skipping in the gastrocnemius muscle of reporter mice following dual gRNA delivery [202]. The potential of NanoMEDIC was further demonstrated in a parasitic infection model, where inducible heterodimerization between FRB-Cas9 and FKBP12-Gag enabled efficient encapsulation of SpCas9 into EVs, resulting in a significant reduction (approximately 25%) in dnase2 transcript levels as measured by RT-qPCR [203]. Subsequent innovations have refined Gag-based engineering for therapeutic purposes. In one approach, EVs were developed by adding multiple nuclear export signal (NES) and nuclear localization signal (NLS) sequences to the Gag protein, with Cas9 fused to the C-terminal end of the NLSs, thereby optimizing its intracellular trafficking [204].

Collectively, VSV-G- and Gag-based loading strategies represent two paradigms for enhancing EV cargo incorporation and functional delivery. Gag-driven loading primarily exploits its intrinsic self-assembly and budding properties to achieve high-density encapsulation of protein cargos, while VSV-G-mediated loading critically enhances post-loading performance by promoting vesicle fusion, endosomal escape, and cytosolic release of the delivered cargo. Together, these strategies highlight a division of labor between loading capacity and functional delivery, underscoring how distinct viral-derived mechanisms can be selectively harnessed or combined to optimize EV-based protein and gene-editing delivery platforms.

Targeting

TSPANs

CD63

Except for enhancing cargo-loading capacity, engineering CD63 can also improve the targeting specificity of EVs. Targeting strategies for engineered EVs are most commonly achieved through surface display of targeting ligands or peptides, including antibody-based ligands such as human leukocyte antigen-G-VHH antibody-modified (anti-HLA-G) antibodies for the treatment of solid tumors [205], as well as targeting peptides such as RGD and rabies virus glycoprotein (RVG), which mediate receptor-specific interactions with target cells [206, 207].

Similar to anti-HLA-G antibodies, surface-displayed fibroblast growth factor 21 (FGF21) binds FGF receptors and the β-klotho (FGFR-β-Klotho) on hepatocytes to induce receptor-dependent functional effects [208]. For targeting peptides, EVs carrying miR-26a and engineered by fusing CD63 with Apo-A1 can be specifically delivered to scavenger receptor class B type 1-expressing liver cancer cells, thereby reducing their proliferation and migration rates [209]. Then, surface display of thrombopoietin (TPO)-mimic peptides or GE11 peptides on EVs enables specific targeting of c-Mpl⁺ AML cells [210] and epidermal growth factor receptor (EGFR)-positive tumor cells [211]. Beyond targeting peptides, another class of EV surface engineering strategies involves glycoengineering, co-expression of Fucosyltransferase 7 with CD63-GD-NanoLuciferase (NLuc) enabled the display of sialyl Lewis X (sLeX) motifs on EV surfaces, thereby enhancing binding to colon epithelial cells in a colitis mouse model [212]. Beyond direct display of peptides or glycan motifs, a modular surface engineering strategy has been developed. An Fc-binding domain (such as the Z domain of protein A) was linked to the C-terminus of CD63, creating Fc-EVs capable of binding IgG antibodies. Decorating these vesicles with HER2-targeting antibodies (trastuzumab) or PD-L1 antibodies (atezolizumab) enhanced uptake in HER2-positive breast cancer and PD-L1-positive melanoma cells, respectively. PD-L1 antibody-modified Fc-EVs displayed significantly higher accumulation in tumor tissue in B16F10 melanoma models [53].

CD63-mediated EV surface display can be applied in antiviral studies and vaccine applications. Anti-SARS-CoV-2 nanobodies can be displayed on the EV surface via CD63, enabling binding to the SARS-CoV-2 spike protein and functional neutralization of the virus in vitro [213]. Currently, the EV surface display strategies have also been explored in vaccine development. A novel DNA vaccine (pCSP) employed CD63 to incorporate the antigenic fragments into EVs, and displayed the SARS-CoV-2 receptor-binding domain (RBD) fragment on the EV surface, thereby enhancing the DCs uptake. Activated DCs presented the antigens to cytotoxic T lymphocytes, which then migrate to tumor sites and eliminate tumor cells [214].

CP05 is a short peptide ligand identified through phage display screening and acts as an intermediary binding module that enables the attachment of targeting molecules to EV scaffold proteins. The concept of CP05 as a universal EV anchoring peptide was first proposed by Gao et al. Surface-displayed CP05 binds to CD63, enabling phosphorodiamidate morpholino oligomer (PMO) to be anchored onto EVs, while a muscle-targeting peptide (M12) was introduced, resulting in significant restoration of dystrophin expression and improved muscle function [215].

CD63-mediated targeting exemplifies a general surface-engineering strategy for directing interactions between EVs and specific cellular receptors and tissue microenvironments. Rather than relying on intrinsic tropism, CD63 functions as a programmable anchoring module that enables the surface display of targeting moieties on EVs. Consequently, CD63-based designs provide a foundational framework for EV targeting, supporting the development of targeted EV delivery systems across diverse disease contexts.

CD9

Engineering strategies to enhance the targeting capability of CD9 primarily rely on the surface display of functional molecules. Similar to CD63, CD9 can also serve as an engineering scaffold for EV surface modification to display targeting ligands, thereby enhancing the targeted delivery capability of EVs. For example, insertion of the RGD peptide into CD9 enables its surface presentation on EVs, which markedly enhances the interaction between EVs and αvβ3 integrins on pancreatic ductal adenocarcinoma (PDAC) cells. As a result, EV cellular uptake is increased in both in vitro and in vivo models, and the therapeutic efficacy is further improved [216]. For CD9, given that angiotensin-converting enzyme 2 (ACE2) serves as a key receptor for SARS-CoV-2 entry [217], truncated CD9 variants have been engineered to display soluble ACE2 on EV surfaces, functioning as decoy receptors that block viral spike protein binding and prevent infection, thereby providing a potential EV-based antiviral platform [218]. Similarly, genetic fusion of CD80 to the CD9 scaffold allows CD80 to be presented on EVs. CD80 can bind CD28 on T cells and provide costimulatory signals, thereby significantly promoting the proliferation of antigen-specific CD4⁺ and CD8⁺ T cells [219, 220]. In addition, EVs were designed to co-display anti-CD3 and anti-EGFR single-chain variable fragment (scFv) anchored via the PDGFR TMD, together with the immune checkpoint modulators PD-1 and OX40L displayed through CD9 fusion on the EV surface. The αCD3-αEGFR-PD-1-OX40L EVs achieved robust growth inhibition and elicited robust cellular immunity, whereas EVs displaying only targeting antibodies showed markedly reduced anti-tumor efficacy in an aggressive CRC model. The engineered EVs significantly inhibited tumor growth, increased intratumoral CD8⁺ T-cell infiltration, and reduced Tregs in the tumor microenvironment [221, 222].

CD81

CD81 has been investigated as a scaffold protein to enhance the targeting capacity of EVs, typically through genetic fusion of targeting peptides to CD81. Displaying CD19 scFv on the EV surface using CD81 as a membrane-anchoring scaffold significantly increases the abundance of targeting modules on EVs and results in strong cytotoxic activity against CD19-positive target cell models [223]. Similarly, by introducing the PDGFRβ-TM at the N-terminus of CD81 and genetically fusing Protein A for EV surface display, these engineered EVs showed an approximately 1.4-fold increase in EV-cell binding signals in HEK293 cells stably overexpressing GLP1R compared with control EVs [224]. In addition to displaying targeting molecules on the EV surface through fusion expression, structural engineering of the CD81 has been explored. For example, disulfide bonds were introduced into the large extracellular loop of CD81 to enhance structural stability, followed by random mutagenesis at specific positions within CD81-LEL and subsequent selection of CD81-LEL variants, thereby enhancing antigen-dependent cellular uptake of engineered EVs [225].

These studies emphasize CD81 as a highly engineerable EV scaffold that supports both direct fusion-based targeting and scaffold-centric structural optimization, thereby strengthening functional targeting outcomes.

Other TSPANs

Beyond their role as efficient scaffolds for cargo loading, these TSPANs also provide a versatile platform for engineering EV targeting through surface modification. After displaying sLeX on the EV surface via TSPAN2 and TSPAN3, the uptake of engineered EVs by activated endothelial cells was significantly enhanced [93]. However, studies investigating the engineering of TSPANs to enhance EV targeting remain relatively limited compared with other commonly used TSPANs such as CD63. Therefore, further investigations are required, for example, to determine whether conjugation of targeting peptides to the surface of TSPAN2 or TSPAN3 can improve EV targeting capability.

Type I transmembrane protein

Lamp2b

Lamp2-mediated surface display enhances the targeting capability of EVs and improves the delivery efficiency of cargo molecules, thereby providing novel therapeutic strategies across a wide range of disease contexts. For example, it can be applied to the treatment of neurological disorders, including neurodegenerative diseases [226, 227], spinal cord injury [228], and ischemic stroke [229]. It can also be used for respiratory diseases, such as acute lung injury [230, 231], as well as cancer therapy [232–236].A major application of Lamp2b engineering is the surface display of targeting peptides. Using this strategy, various targeting peptides have been applied to treat different diseases. For example, the chondrocyte-affinity peptide (CAP) has been used to confer chondrocyte targeting, enabling engineered EVs to achieve therapeutic effects in osteoarthritis [237–239]. In addition, the display of different targeting peptides can be leveraged to treat diseases such as acute MI and kidney disease-associated complications [240, 241]. Lamp2b-mediated EV surface display has been well validated, its engineering efficiency can be further improved through molecular-level optimization, such as the introduction of glycosylation motifs to enable robust display of targeting peptides on the EV surface [242].

Among all Lamp2b-based targeting strategies, RVG-modified EVs represent the most extensively validated platform for central nervous system delivery. Initial comparisons showed that RVG-modified EVs accumulated preferentially in organs rich in acetylcholine receptors compared with unmodified vesicles [243], and this targeting strategy has been applied to EVs carrying diverse cargo, including RNA, proteins, and small-molecule therapeutics [244, 245]. For example, displaying RVG on the surface of Lamp2b and loading brain-derived neurotrophic factor (BDNF) inside EVs markedly increased BDNF levels in key brain regions, alleviated depression-like behaviors, and effectively reduced neuroinflammation [246]. More recent approaches introduced RVG-Lamp2b cassettes into the adeno-associated virus integration site 1 (AAVS1) locus of human induced pluripotent stem cells using CRISPR-Cas9, resulting in stable production of EVs with enhanced neuronal uptake mediated through nicotinic acetylcholine receptors, thereby enhancing the brain-targeting capability. Specifically, fluorescence intensity in the brains of mice treated with RVG-modified EVs was more than 1.5-fold higher than that observed in mice treated with control EVs [247]. In addition to protein cargos, RVG-engineered EVs can also be loaded with small-molecule drugs to achieve disease mitigation and therapeutic effects. For example, RVG-EVs loaded with bumetanide significantly alleviated CSDS-induced depression-like behaviors and improved chloride homeostasis in the dorsal hippocampus, thereby restoring GABA-mediated inhibitory neurotransmission [248]. RVG-engineered EVs are most commonly combined with RNA delivery, particularly miRNAs and siRNAs. Functional applications of EVs with RVG-Lamp2b include delivery of miR-124 to promote neurogenesis and reduce brain injury [249], and suppression of cocaine-induced microglial activation [250]. Engineered RVG-EVs loaded with miR-29 accumulated in the unilateral ureteral obstruction (UUO) kidney and prevented UUO-induced body weight loss and muscle atrophy, supporting their potential as a therapeutic strategy for UUO-associated disease [251], while RVG-EVs carrying miR-137 alleviated autism-like behaviors [252]. In addition, RVG-engineered EVs have been widely explored as siRNA delivery vehicles for the treatment of CNS disorders. In early studies of RVG-EV-mediated siRNA delivery, systemic administration of targeted EVs was shown to achieve up to 60% RNA and protein knockdown, predominantly in the midbrain, cortex, and striatum of mice [253], as well as BACE1 mRNA level decreased 61%±13% and β-amyloid the total amount of β-amyloid decreased by 55% [86]. RVG-EV-delivered opioid receptor mu (MOR) siRNA potently suppressed morphine relapse by downregulating MOR expression [254], while delivery of HMGB1 siRNA markedly reduced HMGB1 expression and infarct volume, indicating their potential as a therapeutic delivery system for ischemic stroke [255]. Moreover, in vivo-assembled RVG-sEV-siRNA successfully crossed the BBB, silenced mutant superoxide dismutase type 1 (SOD1) in the CNS, and significantly alleviated amyotrophic lateral sclerosis (ALS) symptoms [256]. Engineered EV-mediated siRNA delivery has emerged as an important therapeutic strategy for neurodegenerative diseases. For example, RVG-Lamp2b EVs can also be used to load and deliver siRNA to the brain, achieving approximately 40% silencing of the Htt gene in the mouse cortex and a highly significant reduction in p62-positive inclusion bodies in cortical neurons [257]. The delivery strategy has also been further validated in other studies [258]. This indicates that Lamp2b-based EVs displaying targeting peptides also possess therapeutic potential for neurodegenerative diseases.

RGD (Arg-Gly-Asp) is a classical integrin-binding motif that can recognize and bind integrins such as αvβ3 and αvβ5. Because these integrins are highly expressed on tumor vascular endothelial cells and various tumor cells, RGD peptides are commonly used to achieve integrin-mediated tumor-targeted delivery [259]. For example, engineering of EVs can exploit the blood-spinal cord barrier-penetrating capability of Angiopep-2 (Ang2) together with the pathological neovascular targeting property of RGD to achieve precise localization of EVs to spinal cord injury sites [260]. Furthermore, targeted delivery of perlecan by RGD-M2-sEVs promotes BSCB repair and offers a potential therapeutic approach for spinal cord injury [261]. Similarly, RGD-EVs exhibited approximately 40% higher accumulation in glioblastoma (GBM) cells than native EVs. When loaded with doxorubicin, RGD-EVs showed an approximately 13% stronger cytotoxic effect on HROG36 and U87 MG cells compared with doxorubicin delivered by native EVs [262]. While RGD primarily enhances integrin-mediated tumor targeting, internalizing RGD peptide (iRGD) further enables deep tumor penetration through activation of the CendR-neuropilin-1 pathway. Accordingly, iRGD is also a suitable ligand for surface display on engineered EVs, enabling their application in the treatment of breast cancer [263]. The fusion of αv integrin-binding iRGD peptides with Lamp2b in immature dendritic cells (DCs) enabled the generation of engineered EVs for doxorubicin delivery, which selectively accumulated in tumor tissue and suppressed tumor growth without overt toxicity. Tumor volumes in the control group increased by approximately 15-fold, whereas tumors treated with iRGD-Exos-Dox showed only an approximately 4-fold increase [264]. Similarly, Lamp2b-mediated co-display of iRGD and tyrosine fragments on the EV surface achieved similar tumor-suppressive effects [265]. Moreover, siRNA delivered by iRGD-engineered EVs exhibited a stronger inhibitory effect on cell proliferation in N87 cells, supporting siHER2-loaded EVs as a potential therapeutic strategy for HER2-positive gastric cancer [266].

Cardiac-targeted EVs represent another prominent line of development, in which the surface display of cardiac-related peptides is a commonly employed approach. Lamp2b-mediated surface display of a cardiomyocyte-binding peptide (CMP) markedly enhanced EV targeting, resulting in an approximately 18-fold increase in cardiomyocyte uptake at 24 h and a concomitant reduction in apoptosis (16% ± 1%) compared with control EVs (21% ± 1%) [267]. Cardiac-targeting peptides (CTPs) displayed on EVs via Lamp2b confer cardiac specificity, with CTP-Lamp2b-engineered EVs showing an approximately 15% increase in heart delivery in mice and supporting co-loading strategies, such as curcumin with miR-144-3p, to enhance therapeutic efficacy in MI [268, 269]. In addition, the use of heart-homing peptide fused to Lamp2b similarly enhanced delivery of therapeutic miRNAs such as miR-148a, achieving effective treatment of cardiac hypertrophy [270].

Beyond these, Lamp2b has been combined with peptides such as Ang2 to promote targeted repair of spinal cord injury and brain inflammation [271]. In research related to the brain, Ang2-Lamp2b-EVs exhibited higher fluorescence intensity in the brain, indicating enhanced brain targeting. Moreover, Ang2-Lamp2b-EVs loaded with siPgam5 significantly reduced the proportion of TUNEL-positive cells and brain edema. This engineered EV delivery system achieved efficient PGAM5 silencing and promoted recovery of neurological motor and cognitive functions [272]. In addition, the dual-peptide designs further increasing delivery efficiency. Ang2 combined with a transcriptional activator peptide enhanced chemotherapy for gliomas [36], while co-loading of Cas9/sgRNA complexes into Ang/TAT-modified EVs enabled gene editing efficiencies exceeding 65% with minimal off-target effects in GBM [273]. Meanwhile, similar strategies involving the surface display of ABPs have also demonstrated therapeutic potential in certain kidney diseases [274].

For proteins and antibodies, Lamp2b-mediated surface display has also been applied in disease treatment. For example, surface display of PD-1 on engineered EVs has been shown to alleviate primary ovarian insufficiency [275]. The incorporation of an anti-CD206 scFv into TMD of Lamp2b enables surface presentation of macrophage-targeting ligands, thereby facilitating efficient CAR mRNA delivery and resulting in CAR protein expression in approximately 28% of lung macrophages in vivo [121]. In another example, fusion of interleukin-3 (IL-3) with Lamp2b enabled the generation of engineered EVs capable of targeting chronic myelogenous leukemia cells, thereby facilitating efficient imatinib delivery, achieving comparable in vitro efficacy at imatinib doses approximately 37-fold lower than those required for free drug treatment, and suppressing tumor cell proliferation [96], comparable strategies have since been applied to other malignancies, including GBM, thyroid cancer [276]. Similarly, surface display of IL-16 on EVs enables specific targeting of CD4⁺ T cells, thereby facilitating the elimination of latent HIV-1 reservoirs [277]. Moreover, the fusion of Lamp2b with ephrin-B2 has been shown to enhance the tissue-targeting ability of engineered EVs, thereby enabling more effective treatment of ovarian cancer [147].

Collectively, with respect to cargo loading, Lamp2b has been investigated more extensively for its targeting capability (Fig. 4). Owing to its favorable membrane topology, Lamp2b is particularly well suited for the surface display of targeting peptides that mediate receptor-dependent recognition and uptake in specific tissues. Consequently, Lamp2b-mediated targeting has become a foundational approach for the construction of targeted EV delivery systems.

Fig. 4.

Functional classification of scaffold proteins used in EV engineering. (a) Functional classification of scaffold proteins. Scaffold proteins are used for targeting, loading, both targeting and loading, or surface display without targeting. Targeting accounts for 50% (N = 60), loading for 32% (N = 91), scaffolds supporting both targeting and loading for 14% (N = 40), and displaying but not targeting for 4% (N = 12); percentages and corresponding study counts are indicated. (b) Distribution of scaffold proteins used for targeting. Targeting scaffolds are dominated by type I transmembrane proteins, with Lamp2b (39%, N = 60) and PDGFR (12%, N = 19) accounting for the relatively large proportion, followed by TSPANs (e.g., CD63, CD9, CD81). (c) Distribution of scaffold proteins used for both targeting and loading. CD63 is the most commonly used scaffold (35%, N = 9), followed by Lamp2b (19%, N = 5), PTGFRN (12%, N = 3), and C1C2 (12%, N = 3) with several other scaffolds reported at lower frequencies. (d) Distribution of scaffold proteins used for loading. TSPANs are the most frequently used scaffold protein for cargo loading, with CD63 being the most commonly used scaffold (23%, N = 31), followed by CD9 (14%, N = 19) and CD81 (10%, N = 13). In addition, multiple other scaffold proteins have been explored for cargo loading

PDGFR

Engineering strategies targeting PDGFR have predominantly focused on the surface display of targeting peptides, thereby conferring specific targeting capability to engineered EVs.

Antibody-based ligands, such as full-length antibodies and antibody-derived scFvs, have been extensively employed to confer targeting specificity to engineered EVs. For example, surface display of anti-CD2 scFv on EVs resulted in over a 100-fold increase in binding to CD2⁺ Jurkat T cells compared with non-targeted EVs [278]. Likewise, by stably displaying IL-15 and a cetuximab scFv on NK cell-derived sEVs, reducing tumor growth by over 42% and tumor weight by approximately 30% compared with controls [279], and the display of death receptor 5 (DR5) agonistic scFvs resulted in approximately 50% of vesicles having detectable DR5-scFvs and enabled potent cytotoxic activity against DR5-positive tumor cells [280]. Both strategies effectively suppressed tumor progression. Similarly, display of anti-PD-L1 scFvs on the EV surface has also been explored for cancer immunotherapy [281, 282]. For antibodies, a representative application is the development of scaffold-mediated antigen receptor targeted EVs, a synthetic multivalent antibody redirection platform in which anti-human CD3 and anti-human epidermal growth factor receptor 2 (anti-HER2) antibodies are displayed on the EV surface. These engineered vesicles are capable of simultaneously engaging T-cell CD3 and the HER2 receptor on breast cancer cells, thereby redirecting cytotoxic T cells to tumor sites to attack breast cancer cells. In preclinical models, the engineered EVs exhibited strong and specific antitumor activity both in vitro and in vivo [283]. In parallel, through surface display of the anti-Trop2 nanobody or anti-transferrin receptor 1 antibody on EVs, engineered EVs can be developed for targeted cancer therapy [284, 285]. In addition to antibody-based ligands, surface display of cytokines such as IL-2 has also been explored for cancer therapy [286]. Overall, this further demonstrates the multifunctional versatility of PDGFR as a general-purpose scaffold for EV surface engineering in cancer therapy.

Peptide-based targeting displayed on the EV surface via the PDGFR TMD also represents an engineering strategy for EV modification. Through PDGFR-mediated EV surface display of targeting peptide, engineered EVs are precisely localized to specific target cells, thereby enabling the functional effector molecules to induce target cell killing [287, 288]. Meanwhile, display of the p51 peptide on the EV surface enables HER2 targeting. These engineered EVs reduce the relative density of PD-L1 and result in the smallest tumor size as well as lowest tumor weight [289], and surface display of a targeting peptide enhances the delivery of miR-150 to activated hepatic stellate cells, thereby alleviating liver fibrosis [290]. Similarly, PDGFR has also been exploited as an EV surface scaffold to display neurotropic peptides such as RVG, thereby extending PDGFR-based engineering strategies toward brain-targeted delivery. For example, RVG-modified EV-AAV exhibits enhanced brain-targeting capability. RVG-EV-AAV showed approximately 5-fold higher luciferase activity in the striatum and brain-to-peripheral transduction ratio compared with untargeted EV-AAV [291]. Similarly, Henriques et al. demonstrated that RVG-EV-AAVs showed significantly higher bioluminescence in the head and MJD mice treated with engineered vesicles loaded with miR-ATXN3 exhibited a significantly reduced hindbase width and preserved cerebellar interlobular thickness [292].

Collectively, similar to Lamp2b, engineering efforts involving PDGFR have focused predominantly on its targeting function rather than its cargo-loading capacity. Notably, most studies do not exploit the full-length PDGFR receptor, but instead rely on its TMD as a minimal membrane-anchoring module to enable surface display of targeting ligands.

PTGFRN

PTGFRN is also an effective scaffold protein that can be used for EV engineering to enhance targeting capability. For example, by fusion to full-length (FL) or Δ687 PTGFRN, EVs decorated with anti-CD3 scFab induced downregulation of T cell receptor expression on primary mouse CD4⁺ T cells [159]. In addition, fusion of truncated PD-1 to PTGFRN enabled the generation of engineered EVs that significantly reduced tumor volume and prolonged survival (50–60 days), and also markedly increased CD8⁺ T-cell infiltration (36% vs. 23% and 17% in control groups) as well as elevated the levels of pro-inflammatory cytokines IFN-γ and TNF-α [293]. For PTGFRN, an Fc/Spa interaction system was designed by fusing the human Fc domain with the intracellular region of PTGFRN-Δ687 and anchoring it to the EV membrane, while additional modification with an RVG peptide conferred neurotropism and facilitated EV accumulation in neural tissues. RVG-modified EVs achieved a higher indel mutation frequency (25%) compared with non-targeted EVs (20%) and resulted in an 80% survival rate in HSV1-infected mice, indicating potential therapeutic advantages for HSV1 infection and other neurological disorders [161]. In addition, surface display of IL-12 on EVs markedly enhances antitumor therapeutic efficacy, further demonstrating the multifunctionality and extensibility of PTGFRN as a scaffold for EV surface engineering of functional molecules [99]. Moreover, PTGFRN-mediated surface display of EGF-A enables engineered EVs to efficiently sequester PCSK9 and deliver it to lysosomes for degradation, thereby restoring endogenous Ldlr levels and alleviating atherosclerosis [294].

Lipid-Anchored membrane proteins

GPI

GPI-anchored proteins represent a class of membrane-associated proteins that are tethered to the outer leaflet of the plasma membrane through a GPI anchor. This anchoring mechanism has been exploited in EV engineering to enable the surface display of diverse functional proteins. The incorporation of GPI-anchored nanobodies on EVs has emerged as a novel strategy to extend their functional repertoire, facilitating the presentation of antibodies, reporter proteins, and signaling molecules. For example, by displaying anti-EGFR nanobodies on the EV surface via GPI anchoring, the engineered EVs showed more than a 10-fold higher association with A431 cells compared with control EVs [295]. Notably, the efficiency of this strategy is also strongly influenced by the density of EGFR on recipient cells [296]. Similarly, anti-GD2 nanobodies were displayed on the EV surface via GPI anchoring, resulting in enhanced tumor accumulation, with the fluorescence intensity of GD2-engineered EV mimetics being higher than that of control EV mimetics at 48 h [297]. Beyond direct nanobody display, a GPI-anchored adapter was engineered by fusing streptavidin, NLuc, and mCherry, enabling its stable display on the EV surface. Through streptavidin-biotin interactions, biotinylated anti-CD8 antibodies were recruited onto the EV surface, thereby conferring CD8-specific targeting. This strategy resulted in up to a 43-fold increase in NanoLuc luminescence signal in CD8-GFP-expressing cells, reflecting markedly enhanced EV binding and uptake [298]. In addition, insulin surface display mediates targeted EV delivery, and such engineered EVs significantly suppress cardiomyocyte apoptosis under ischemia-reperfusion conditions, reducing the apoptotic rate from 15% to 6% [299].

Beyond these, GPI-anchored proteins have also been leveraged for immunomodulatory applications. Incorporation of CD14, a prototypical GPI-anchored receptor, onto the surface of EVs resulted in a striking reduction of lipopolysaccharide (LPS)-induced secretion of tumor necrosis factor-α and IL-6. Mechanistic analysis suggested that these engineered EVs could capture LPS, thereby limiting its interaction with Toll-like receptors on macrophages and ultimately achieving LPS neutralization [300]. These findings highlight the versatility of GPI-anchored proteins in expanding the targeting capacity and functional potential of EVs, with implications for both cancer therapy and inflammatory disease modulation.

GPI anchoring offers a topologically distinct targeting strategy characterized by outward-facing ligand display, high surface accessibility, and minimal coupling to EV biogenesis, making it particularly attractive for antibody- and nanobody-mediated targeting applications.

Peripheral membrane proteins

C1C2

Studies have demonstrated that the C1C2 fusion strategy can simultaneously provide EVs with targeting and delivery functions. For example, by fusing EGFP to the C1C2 domain of lactadherin, EGFP was successfully displayed on the surface of EVs. This EV display strategy significantly enhanced antigen-specific humoral immune responses following both intramuscular and intranasal vaccination, thereby improving the immunogenic potency of ChAdOx1 and Ad5 vaccine vectors [301]. In another study, anchoring the EBV-derived CD21 ligand D123 to the EV surface enabled efficient delivery of the OVA antigen to B cells and significantly enhanced antigen-specific CD8⁺ T cell responses, accompanied by an approximately 5-fold increase in in vitro B cell targeting compared with control EVs [302].

The C1C2 domain has been widely used to display monobodies on EV surfaces. Using this strategy, monobody repertoires can be presented on EVs to enable in vivo selection of tissue-targeting binders [303]. Following library-based discovery, individual monobodies, such as the EGFR-targeting monobody E626, were further displayed on EV surfaces to achieve receptor-specific targeting. E626-displaying EVs showed rapid association with A431 cells, with detectable binding observed as early as 10 min after incubation, indicating that the E626 monobody effectively enhances EV binding to target cells [182]. In addition, the display of targeting peptides represents an effective strategy. For example, EVs displaying the ischemia-homing peptide on their surface via the C1C2 domain exhibited significantly enhanced uptake in ischemic neonatal rat ventricular myocytes (NRVMs) compared with non-targeting control EVs. Moreover, this ischemia-homing peptide surface display markedly enhanced EV localization and delivery to injured myocardial tissue following simple intravenous (IV) administration, rather than merely causing nonspecific accumulation in heart tissue [52]. Then, by fusion the p88 to C1C2, NIT-1 cells exhibited stronger bioluminescence after 30 min of incubation with p88-gLuc-EVs compared with cells treated with NP-gLuc-EVs, demonstrating an enhanced binding capacity of p88-displaying EVs toward β-cells relative to non-peptide EVs [304]. Collectively, these findings highlight the potential of C1C2 as a versatile scaffold module for EV engineering.

Syntenin-1

Syntenin-1 acts as an adaptor and scaffold protein through its two PSD-95, Dlg, and ZO-1 (PDZ) domains, enabling its participation in multiple signaling pathways and modulation of cellular physiology. In addition, syntenin-1 is closely associated with EV biogenesis, where it regulates the production and release of EVs by coordinating endosomal sorting processes [305].

Syntenin-1 has been exploited as a scaffold protein to display decoy receptors on EVs, thereby enabling therapeutic intervention in disease models. For example, by fusing tumor necrosis factor receptor 1 (TNFR1) and IL-6 signal transducer (IL6ST) to the N-terminal sorting domain of syntenin-1, engineered EVs were generated to display cytokine decoy receptors on their surface, which can act as decoys for the pro-inflammatory cytokines TNFα and IL-6. In a TNBS-induced colitis model, survival was significantly improved by double-decoy EV treatment (92%) compared with untreated mice (67%) [306]. Similarly, syntenin-mediated display of the decoy receptor TACI on EVs enabled binding of the cytokine B lymphocyte stimulator (BlyS) and attenuated systemic lupus erythematosus (SLE) severity, markedly reducing inflammatory kidney damage in SLE mice by blocking BLyS/APRIL signaling [307].

Collectively, although both C1C2 and syntenin-1 are peripheral membrane-associated proteins, their engineering strategies for EV targeting diverge markedly in practice. C1C2-based designs are predominantly used to display targeting ligands or peptides on the EV surface, reflecting its suitability as a simple anchoring module for outward-facing functional motifs. In contrast, engineering efforts involving syntenin-1 have largely focused on the presentation of decoy receptors rather than classical targeting ligands. This may be attributed to the cytosolic localization of syntenin-1, which restricts its role to enriching proteins that are natively oriented toward the extracellular space, while decoy receptors have substantially lower requirements for precise outward topology than targeting peptides. Together, these patterns highlight how the intrinsic functional context of peripheral membrane proteins shapes their practical use in EV surface engineering.

Virus related proteins

VSV-G

In addition to facilitating immune evasion and serving as a scaffold protein for EV engineering to enhance cargo loading capacity, VSV-G also plays a role in improving the targeting properties of engineered EVs. The successful display of the S1 antigen via the TMD of VSV-G validates the feasibility of using VSV-G as a transmembrane anchoring module for molecular surface display [308]. By genetically fusing the RBD to VSV-G and displaying it on the EV surface, engineered EVs delivered siRNA to ACE2-high-expressing recipient cells, resulting in a significant reduction of GFP signals in the lung (about 50% decrease) [128]. Similarly, truncated VSV-G has been employed to enable surface display of anti-CD206 nanobodies, thereby facilitating targeted delivery of cargo molecules [309].

In summary, VSV-G has been shown not only to possess fusogenic activity and cargo-loading capability but also to function as a scaffold for mediating the surface display of targeting molecules on EVs, further demonstrating its effectiveness as an EV scaffold protein.

Stability

For stability, CD47 is a ubiquitously expressed cell surface protein belonging to the Ig superfamily. It is heavily glycosylated and structurally characterized by an extracellular Ig variable domain, five transmembrane segments, and a short cytoplasmic tail that undergoes selective splicing [310]. Functionally, CD47 interacts with receptors such as SIRPα on immune cells to inhibit phagocytosis, thereby serving as a key regulator of immune evasion. Evidence from breast cancer-derived EVs has demonstrated that CD47 engagement with SIRPα modulates signaling pathways in endothelial cells, influencing phosphorylation of vascular endothelial growth factor receptor 2 (VEGFR2) and altering the expression of angiogenesis-associated genes [311]. This highlights a role for CD47 in linking immune checkpoint regulation with vascular remodeling. In addition, red blood cell-derived EVs have been engineered to exploit the intrinsic immune-modulatory properties of CD47. By the cell membrane properties of CD47, the EVs can evade clearance by the immune system, thereby achieving a longer circulation time in vivo [312].

Introducing the albumin-binding domain (ABD) markedly improved the circulatory stability of engineered EVs, enhancing their systemic persistence. For example, incorporation of the ABD into the LEL of TSPANs, followed by addition of NLuc to the C-terminus for quantification, demonstrated that ABD-displaying EVs achieved markedly higher plasma concentrations compared with wild-type controls, and TSPAN2-ABD-NLuc resulted in up to a 12-fold increase in plasma concentration at 60 min compared to TSPAN2-NLuc [93]. CD63-engineered EVs exhibited more than a 10-fold increase in circulating concentrations at about 5 h post-injection compared with control EVs, indicating enhanced plasma retention [313].

Except for ABD, the display of other specific peptides can also improve stability of engineered EVs. For example, Lamp2b-mediated display of CMP significantly increased the cardiac retention of cardiosphere-derived cells (CDC)-derived EVs, with cardiac biofluorescence reaching 4 ± 1 compared with 2 ± 1 for control EVs [267]. Surface display of ABP via Lamp2b resulted in significantly higher in vivo signals of engineered EVs at 2, 4, and 8 h compared with control EVs, indicating prolonged circulation of ABP-EVs [274]. Fusion of a collagen-binding domain (CBD) peptide to Lamp2b enabled stable tethering of EVs to collagen I (Col-I) scaffolds, thereby enhancing the retention of miR-21-loaded EVs at lesion sites [228]. sLeX was displayed on the EV surface via engineering the large extracellular loop of the scaffold protein CD63, which resulted in increased plasma retention of EVs, with F7CCN showing a 1.8-fold higher level than non-targeting EVs [212]. In addition, by fusing the tissue-homing peptide to the N-terminus of CD64 to create a “CD64-CK” fusion protein, serving as a universal anchoring platform on EVs for binding humanized monoclonal antibodies for targeted delivery. This approach significantly inhibited growth and metastasis of pancreatic carcinoma cell line 1 tumors and patient-derived xenograft in mice, while also prolonging survival in mice [63] Anchoring PEG onto the EV surface via the CP05 peptide forms a hydrophilic shielding corona, enabling engineered EVs to effectively escape phagocytosis by RAW264.7 macrophages and non-target organs [314]. EVs decorated with PEG2000-TK-CP05 efficiently evade mononuclear phagocyte system (MPS)-mediated clearance, thereby prolonging their circulation time in the bloodstream [315].

Overall, enhancement of the stability of engineered EVs primarily relies on the introduction of functional modules on the EV surface that confer immune evasion, plasma protein binding, or tissue retention. In addition to suppressing phagocytic clearance through the “don’t-eat-me” signal mediated by CD47, the display of ABD or other specific peptides on the extracellular domains of scaffold proteins can also promote EV retention at specific sites. Similar to targeting strategies, these stability-enhancing approaches fundamentally act by modulating the interactions between EVs and the host immune system or tissue microenvironments through surface engineering.

Summary of scaffold proteins in EV engineering

Taken together, scaffold proteins provide the two principal design purposes for EV engineering: cargo loading and targeted delivery. Structurally, commonly used scaffolds fall into several major classes, including TSPANs (e.g., CD63, CD9), type I single-pass transmembrane proteins (e.g., Lamp2b, PDGFR, Plexin, PTGFRN), lipid-anchored or peripheral membrane modules (e.g., GPI anchors, N-Myr tags, BASP1, C1C2, ARRDC1), and virus-related components (e.g., VSV-G, Gag). These scaffolds differ substantially in membrane topology, EV-sorting mechanisms, and intracellular trafficking routes, leading to distinct functional biases in EV engineering applications.

Functionally, scaffold protein-based EV engineering has been investigated across a broad range of therapeutic indications, including oncology, cardiovascular disease, neurodegeneration, and inflammatory or antiviral disorders. As summarized in Figs. 4 and 5, distinct scaffold classes show clear functional biases toward specific engineering objectives and cargo modalities. TSPAN-based scaffolds (CD63, CD9, CD81) are most frequently employed for intraluminal loading of proteins, nucleic acids, and CRISPR components, reflecting their robust EV-sorting capacity and broad compatibility across producer cell types (Fig. 4). In contrast, Lamp2b-based strategies are predominantly applied for surface display of targeting ligands, enabling tissue- or organ-specific delivery through peptides such as RVG, iRGD, or cardiac-targeting motifs, and are therefore widely used in CNS, tumor, and cardiovascular models (Fig. 4; Table 1). Type I transmembrane scaffolds, including PDGFR and PTGFRN, support multivalent surface display and high-density cargo packaging, whereas lipid-anchored or peripheral scaffolds (e.g., BASP1 and ARRDC1) provide alternative routes for dense protein or RNA loading and, in some cases, plasma membrane-derived vesicle biogenesis. Virus-related scaffolds such as Gag and VSV-G are preferentially used for genome-editing payloads, where enhanced membrane fusion or cytosolic access is required, albeit with additional translational and safety considerations (Fig. 5; Table 2).

Fig. 5.

Overview of scaffold protein-mediated EV cargo types. (a) Overall distribution of cargo modalities. Scaffold protein-based EV engineering strategies enable loading of protein, nucleic acid, and RNP. Protein cargos are the most common category (61%, N = 66), followed by RNP cargos (24%, N = 26), and nucleic acid cargos (16%, N = 17); percentages and corresponding study counts are indicated. (b) Distribution of scaffold proteins used for protein loading. TSPANs are the most frequently used scaffolds (e.g., CD63, CD9, CD81), together with type I transmembrane proteins, virus-related proteins, and other less frequently employed scaffolds. (c) Distribution of scaffold proteins used for nucleic acid loading. The selection of scaffolds is more restricted, with CD63 and Lamp2b being most commonly applied, alongside a limited number of alternative membrane-associated scaffolds. (d) Distribution of scaffold proteins used for RNP loading. No specific scaffold protein class is exclusively preferred, with virus-related proteins (Gag, 25%, N = 8) being the most frequently used individual scaffold

Table 1.

Representative studies of scaffold protein-based EV engineering

Scaffold protein	Cargo	Engineering strategy	Engineering efficiency	Applications	Therapeutic efficiency	Ref
TSPANs
CD63	Fc-binding domains Antibodies: Trastuzumab and Atezolizumab	Genetic fusion Interaction	105 APC-labeled hIgG per vesicle	Cancer immunotherapy	100% survival compared to 35% in control	[53]
	Nanoluc ABDs	Genetic fusion	/	Immunotherapy	> 10×circulating EV levels	[313]
	Cas12a mRNA Anti-HIV crRNAs Nb1 (Anti-CD4 nanobody)	Genetic fusion Interaction	10×intracellular Cas12a expression 15% Nb1 + EVs	HIV-1	28× uptake in CD4 + cells vs. non-targeted EVs 100% survival Plasma p24 decrease CD4/CD8 ratio restored	[122]
	PUFe Cre recombinase mRNA (Nanoluc, mOx40L)	Genetic fusion Interaction	89% bilayer-protected mRNA	Aggressive melanoma	67% complete remission	[120]
	anti-mesothelin CAR mRNA	Genetic fusion Interaction	693 CAR mRNA copies/1 × 106 EVs	in vivo CAR macrophages Lung metastasis and recurrence	28% lung macrophages CAR+	[121]
	Ndufs1 mRNA VegfA mRNA	Genetic fusion Interaction	450 ng Ndufs1 mRNA/mg EVs 240 ng VegfA mRNA/mg EVs	Myocardial Infarction (MI)	Improved cardiac function Reduced infarct size	[123]
	RNPs	Genetic fusion Interaction	2–5× Cas9/ABE enrichment	DMD	In vivo NGS INDEL rates in muscle: Up to 0.2%.	[126]
	CP05-conjugated cargo	Interaction	89% CP05 modification efficiency	DMD	19×dystrophin expression	[215]
CD9	Cas9, sgRNA Cre recombinase; protein	Genetic fusion	/	Alzheimer’s disease	Bace1 mRNA reduced Aβ pathology reduced	[132]
	miR-155; miR-155 inhibitor (antimiR-155); CRISPR/dCas9 mRNA; C/ebpα gRNA	Genetic fusion Interaction	98 miR-155 copies/EV (CD9-HuR) 22 dCas9 mRNA copies/100 EVs (3×AREs)	liver injury/fibrosis	7× miR-155 loading Tnfα ↓, Mcp1 ↓, Col1a1 ↓ in liver	[136]
	Ldlr mRNA	Genetic fusion Interaction	/	Familial hypercholesterolemia (FH)	3-fold higher LDLR protein in mouse liver than control group	[134]
	Cas9 protein sgRNA (LoxP and PCSK9)	Genetic fusion Dimerization	> 20 Cas9 molecules/EV	Functional knock-down of the PCSK9 gene	> 4% indel	[131]
	ABD: ABD035 and ABD094	Genetic fusion	/	Boosting CAR T cell persistence and antitumor efficacy in solid tumors (Neuroblastoma)	100% survival 60% specific lysis vs. 20% control	[288]
CD81	EGF-A of LDLR	Genetic fusion	20% of the bulk isolated EVs were engineered	Atherosclerosis (PCSK9)	Total cholesterol ↓; LDL-C ↓ Aortic root plaque area ↓	[294]
TSPAN2/3	Cre recombinase protein/Super-repressor NF-κB (IkB mutant)	Genetic fusion	TSPAN2-EVs showed 4.6-fold enrichment compared with CD63-EVs	Systemic inflammation	60% survival	[141]
	ThermoLuc, mNeonGreen. HiBiT ABD sLeX	Genetic fusion	TSPAN2 and TSPAN3 outperform CD63	Circulation time Endothelial cell-specific targeting	12× plasma concentration	[93]
Type I transmembran protein
Lamp2b	RVG peptide	Genetic fusion	/	Huntington’s disease (NCT06024265)	50% mHTT reduction in cortex 45% mHTT reduction in striatum Motor performance improved	[258]
	RVG peptide	Genetic fusion	SOD1-siRNA enriched in plasma sEVs	ALS	Grip strength improved; rotarod performance improved Gastrocnemius CSA restored; atrophy markers reduced (Atrogin-1, MuRF-1) SOD1 protein reduced; mRNA reduced (spinal cord L3-L5)	[256]
	iRGD peptide	Genetic fusion	/	Tumor treatment	6× smaller tumor volume	[265]
	Viral macrophage inflammatory protein-II	Genetic fusion	/	Viral myocarditis	> 7% heart accumulation at 6 h 22% Ly6C+ monocytes targeted 27% F4/80 + macrophages targeted 35% M1 (iNOS+) ↓ 70% M2 (CD206+) ↑	[148]
PDGFR	DR5-scFvs	Genetic fusion	50% EVs display DR5-scFvs	Cancer immunotherapy	Greater melanoma growth inhibition Mouse survival prolonged	[280]
	IL-2	Genetic fusion	/	Melanoma	Tumor volume reduced PD-L1 protein reduced > 40%	[286]
PTGFRN	RNPs RVG	Genetic fusion	4× SpCas9 packaging 1% SpCas9 of total EV protein	HSV1 infection	80% survival > 1× brain accumulation 25% indel	[161]
	GFP, MS2-MCP, Chicken ovalbumin (OVA), Cas9 Cytokines: IL-7, IL-12 TNFSF Ligands: CD40L, LIGHT anti-CD3 scFab, scFv, and VHH Factor VIII	Genetic fusion	1000 GFP molecules/EV > 95% of EVs contained the scaffold-GFP fusion Cas9: 9 molecules per EV OVA: 168 molecules per EV	Cancer Immunotherapy (NCT05156229)	Tumor suppression improved	[159]
	Ldlr mRNA Flag peptide	Genetic fusion Interaction	/	Familial Hypercholesterolemia	Higher LDLR expression Atherosclerosis plaque decreased	[160]
Plexin	Proteins: GFP, NanoLuc luciferase, IL-12. Nucleic Acids: NanoLuc mRNA. NbALFA (Nanobody against ALFA tag). 7D12 (Anti-EGFR nanobody) Mouse Interleukin 12 RVG peptide	Genetic fusion	Protein Enrichment: higher compared to Lamp2b and PTGFRN mRNA Enrichment: >10 fold higher than passive loading	Cancer Immunotherapy	Tumor weight reduced	[157]
PTTG1IP	GFP Cre recombinase RNPs anti-HER2 scFv	Genetic fusion	GFP Cargo: 75% of vesicles were GFP+ Loading: double the amount of GFP per vesicle Higher EV-loading capacity compared to CD63	Platform development	46–90% in vivo Cre recombination 98% Cre activation 10% reporter activation Up to 6% endogenous RNF2 editing	[163]
Lipid-anchored membrane protein
N-Myr	Cas9 protein, sgRNA	Genetic fusion	110–285 ng Cas9 protein per 1 × 1010 EVs (46–107 Cas9 molecules per EV). Cas9 accounts for 0.7% of total EV protein 3 sgRNA molecules per 10 EVs Cas9:sgRNA ratio is 256:1	Prostate Cancer	15% gene editing in tumor tissue 31% reduction in C4-2B-EnzR cell proliferation	[168]
BASP1	Minimal RISC complex composed of AGO2 protein and saiRNA	Genetic fusion minRISC components	15 AGO2 molecules per 10 minRISC-EVs (100-fold enrichment compared to wild-type EVs) 12 nM guide RNAs at 1E13 particles/mL. BASP1-N10 loaded higher amounts of AGO2 compared to BASP1-FL	Lung injury and tumor	60% silencing efficiency	[173]
GPI-anchored proteins	Insulin	Genetic fusion	/	Myocardial Ischemia-Reperfusion (I/R) injury	Infarct size reduced	[299]
Peripheral membrane protein
ARRDC1	p53 protein p53 mRNA GFP mRNA RNP	Genetic fusion Interaction	540 cargo protein molecules per ARMM vesicle	Platform development	Apoptosis induced in spleen and thymus	[177]
	GFP, Cre recombinase, ABE8e (Adenine Base Editor 8e) GFP mRNA, sgRNA anti-CD8 scFv mCD8 DARPin GluA4 DARPin	Genetic fusion Interaction	> 50% ARMMs displayed surface-protein+	Platform development	Base Editing (ex vivo): 40–60% A-to-G conversion in CD8 + cells vs. ~10% in CD8- cells. Mouse Spleen (in vivo): 7% of CD8 + splenocytes were positive vs. ~1% of CD8- cells. Mouse Brain (in vivo): Over 50% of PV+ neurons were positive, with almost no expression in non-PV cells.	[178]
C1C2	Monobody repertoires/libraries	Genetic fusion	80% of EVs displayed monobodies 20 monobody molecules per EV	Platform development	4× pancreatic accumulation	[303]
Alix	PD-1 FZD8	Genetic fusion	> 90% of CD63 + EVs being positive for the displayed protein	Metastatic melanoma	Lung tumor burden suppressed CD8 + T cell infiltration increased exhausted T cells reduced Complete tumor growth inhibition	[320]
Syntenin-1	TACI decoy receptor protein	Genetic fusion	TACI + EVs: 10–17%	SLE	Activated B cells reduced (CD69+)	[307]
	TNFR1 IL-6ST IL-23B	Genetic fusion	60% EVs TNFR1+ 600 TNFR1 receptors/EV 23% dual-receptor EVs	Inflammation	Sepsis: 100% survival EAE: Clinical score reduced to 1.7 (TNFR1 EVs) and 1.3 (IL-6ST EVs) Colitis: Survival improved to 92.3%	[306]
Virus-related protein
VSV-G	Cre recombinase, RNPs, Super-repressor of NF-κB activity (SR), Meganuclease targeting PCSK9	Genetic fusion	34 Cas9 molecules/100 EVs (VEDIC) 130 Cas9 molecules/100 EVs (VFIC)	Systemic inflammation Functional delivery to the CNS	40% recombination in hippocampus; 30% in cortex 100% Cre recombination in reporter cells 80% indels at endogenous mTTR	[65]
gag	Cas9 protein sgRNA	Genetic fusion	565 × 108 Cas9-EDV particles/10 cm plate 266 × 109 sgRNA molecules/10 cm plate 330× sgRNA packaging (Cas9-containing particles)	In vivo CAR-T genome editing	> 1% modified alleles in CAR T cells	[204]
	Cas9 ABE8e (Adenine Base Editor) gRNAs (targeting B2M, BCL11A, and HBG1/2)	Genetic fusion	/	Hematological disorders	93% allelic editing in vitro On-target potency increased	[197]
	Cas9 sgRNA Luciferase	Chemical-induced dimerization	35 to 79 active RNP molecules per 10 EVs	DMD	7% genomic deletion > 1% exon skipping > 1% large deletions	[66]
	CRISPR-Cas9, ssODN DNA template	Chemical-induced dimerization	/	Parasitic infection	Reduction in dnase2 transcript levels by 25%	[203]

Table 2.

Conceptual summary of scaffold protein classes for EV engineering

Scaffold class	Typical scaffolds	Engineering objective	Typical cargo	Key application	Major advantages	Major limitations	Translational notes
TSPANs	CD63, CD9, CD81	Loading, targeting	Loading: protein, RNA, RNP Targeting: peptide, protein	Cancer therapy, immunotherapy and genetic disease	Strong endogenous EV enrichment and well-defined topology Versatile for programmable loading and surface display	Cell type- and subpopulation-dependent EV sorting Moderate cargo density and endolysosomal bias	Widely used
Type-I TM proteins	Lamp2b, PTGFRN, PDGFR	Targeting (most), loading	Loading: protein, RNA, RNP Targeting: peptide (most), protein	CNS diseases, cancer, and cardiovascular diseases	Stable membrane anchoring with defined orientation Well suited for ligand/antibody surface presentation	Mainly used for surface display; loading less explored Large size increases engineering complexity	Strong clinical momentum
Peripheral membrane proteins	ARRDC1, C1C2	Loading (most), targeting	Loading: protein, RNP Targeting: protein	Gene therapy and genome editing, inflammatory and autoimmune diseases, cancer targeting and therapy	Efficient ESCRT recruitment and vesicle budding Enable high-efficiency protein and RNP delivery	Generate non-classical EV subtypes (e.g., ARMMs) Limited data on in vivo behavior	Emerging, needs standardization
Lipid-anchored proteins	GPI, N-Myr	Loading, targeting	Loading: protein, RNP Targeting: protein	Gene editing and RNA interference therapy, cancer, inflammatory or autoimmune diseases	Simple, modular membrane tethering strategy	Unstable or reversible membrane association Cargo localization biased to membrane	Emerging, needs standardization
Virus-related proteins	VSV-G, Gag	Loading, entry enhancement	Loading: protein, RNP Targeting: protein	Genome-editing-based gene therapy, antiviral treatment, and cancer	Exceptional cargo loading and delivery efficiency Promote membrane fusion and endosomal escape	High immunogenicity limits repeat dosing Safety and regulatory concerns	Powerful but clinically sensitive

Beyond canonical scaffolds, several newly identified or repurposed proteins further expand the EV engineering toolkit. These include loading-oriented scaffolds such as Wntless [316] and EPN-01 [309, 317], as well as display-oriented platforms such as ClyA [318] and IFITM3 [319]. Recent studies have shown that Alix can serve as a scaffold to mediate the coordinated loading of PD-1 and FZD8, thereby enabling their functional surface display on engineered EVs. This allows the EVs to simultaneously block PD-L1 and Wnt7b signaling pathways and activate T cells, providing a therapeutic strategy against metastatic melanoma [320]. In addition, modularly designed synthetic scaffolds have demonstrated the capacity to load proteins and other cargo types in a programmable manner [97], collectively expanding the structural design space for engineered EVs.

Across reported studies, the selection of scaffold proteins is influenced by the physicochemical and functional requirements of the intended cargo. As illustrated in Fig. 5, protein delivery most commonly relies on TSPAN scaffolds to ensure stable membrane anchoring and preservation of bioactivity (Fig. 5), whereas nucleic acid delivery favors CD63-, CD9-, or Lamp2b-based platforms to facilitate selective encapsulation and intraluminal protection (Fig. 5). By contrast, delivery of RNP complexes-particularly CRISPR-Cas systems-shows a pronounced preference for virus-related or lipid-anchored scaffolds, including Gag and VSV-G, which better accommodate large macromolecular assemblies and promote cytosolic access through enhanced membrane fusion or endosomal escape (Fig. 5). These modality-dependent patterns provide a practical framework for scaffold selection and highlight how EV engineering strategies diverge according to cargo size, stability, and intracellular release requirements.

From a translational perspective, scaffold-engineered EVs have been most extensively explored for mRNA delivery, which constitutes the largest proportion of reported platforms and emphasizes selective RNA loading, protection from degradation, and efficient translation in recipient cells (Fig. 5; Table S1). Protein delivery represents an earlier and conceptually foundational modality, focusing on maintaining cargo functionality and achieving effective intracellular access. More recently, delivery of genome-editing systems-primarily CRISPR-Cas RNPs-has expanded rapidly, driven by the need for transient intracellular exposure, high payload capacity, and controlled cargo release (Fig. 5; Table 1).

Collectively, the representative studies summarized in Tables 1 and 2, together with the comprehensive literature mapping provided in Table S1, demonstrate that distinct scaffold classes are preferentially matched to specific cargo modalities and delivery objectives, supporting the emergence of a modular, “plug-and-play” design paradigm for scaffold-engineered EVs. Importantly, translational progress is no longer confined to preclinical models. As highlighted in Table 1, at least two scaffold protein-engineered EV platforms have advanced into clinical evaluation, including ER2001 (NCT06024265) and the PTGFRN-based exoIL-12 program (NCT05156229), underscoring the growing clinical feasibility of scaffold-based EV delivery strategies despite ongoing challenges in quantitative standardization and cross-study comparability.

EV-mediated delivery of protein via scaffold-protein engineering

Scaffold-protein engineering has enabled EVs to function as programmable carriers for therapeutic protein delivery, overcoming the intrinsic limitations of passive protein encapsulation, including low cargo abundance, heterogeneous loading, and poor functional transfer. By genetically coupling EV-enriched membrane scaffolds with protein cargos or adaptor modules, engineered EVs can actively enrich, display, and deliver functional proteins in vitro and in vivo.

Initial efforts rely on direct genetic fusion of proteins to canonical EV scaffolds such as TSPANs (CD63, CD81, CD9), Lamp2b, or PTGFRN [86, 130, 150]. Although straightforward, these approaches generally result in modest and variable cargo enrichment, reflecting stochastic incorporation into EVs and limited control over cargo orientation. Moreover, irreversible fusion to membrane scaffolds may impair protein function or restrict cytosolic availability after uptake, thereby constraining their effectiveness for intracellular protein therapy.

To overcome these limitations, more recent studies have shifted toward active scaffold-assisted loading strategies, incorporating conditional or high-affinity protein-protein interaction modules. Systems based on optogenetic dimerization [131], chemically inducible dimerization domains [130], or other modular binding pairs enable selective enrichment of soluble protein cargos into EVs without permanent fusion. These designs markedly increase cargo copy numbers per vesicle and, critically, allow delivery of functionally intact proteins such as enzymes, recombinases, or reporters. In both neuronal cultures and in vivo models, scaffold-assisted EVs have achieved efficient cytosolic protein delivery, leading to robust functional readouts including enzymatic activity and genome recombination [150, 159].

Despite these advances, several challenges remain. Scaffold-based EV protein delivery systems frequently exhibit rapid systemic clearance, limited tissue specificity, and sensitivity to cargo size or expression stoichiometry. In addition, increasing molecular complexity raises concerns regarding manufacturing scalability, batch-to-batch consistency, and immunogenicity. Addressing these issues will be essential for translating scaffold-mediated EV protein delivery from proof-of-concept studies toward clinically viable therapeutics.

EV-mediated delivery of CRISPR-Cas systems via scaffold-protein engineering

Efficient EV-mediated delivery of CRISPR-Cas systems relies on active, scaffold-guided cargo recruitment. Early virus-inspired platforms, such as NanoMEDIC and related Gag-based systems, demonstrated that chemically inducible dimerization and RNA packaging signals can co-load Cas9 protein and sgRNA into EV-like particles, enabling functional genome editing in hard-to-transfect cells and in vivo muscle models, albeit with moderate efficiencies and dependence on viral components [66, 202, 203].

Subsequent studies shifted toward EV-enriched or endogenous scaffold proteins, including TSPANs (CD9/CD63/CD81), ARRDC1-derived ARMMs, PTGFRN, and PTTG1IP, to improve cargo loading while reducing reliance on viral capsids. ARRDC1-based EVs and truncated sARRDC1 variants achieved exceptionally high Cas9 payloads—hundreds to about 1000 molecules per vesicle—highlighting the advantage of plasma-membrane budding for large RNP delivery [131, 177, 179]. However, direct fusion of large editors to membrane scaffolds often compromises Cas9 activity or restricts nuclear access, underscoring the need for controlled cargo release.

To balance high loading with functional delivery, multiple groups introduced reversible or tunable interaction modules, including aptamer-protein pairs (Com/com, MS2/MCP), ligand- or light-induced heterodimerization (FKBP/FRB, CRY2/CIBN), and photocleavable linkers (PhoCl, mMaple3). These studies consistently show that maximal genome editing is achieved at intermediate binding affinities, where sufficient loading is retained while allowing efficient dissociation of Cas9 RNPs after EV uptake [119, 126, 132, 135, 167].

Functional delivery remains strongly influenced by endosomal escape and membrane fusion, with VSV-G pseudotyping markedly enhancing editing efficiency across platforms but raising concerns regarding immunogenicity and repeat dosing [167, 169–171]. Targeting ligands and envelope engineering strategies have begun to mitigate these issues by enabling cell- or tissue-restricted genome editing, although they do not recapitulate the potent fusogenic and endosomal escape properties of VSV-G, and in vivo efficiencies remain substantially lower than in vitro benchmarks [161, 197, 204].

Overall, scaffold-protein engineering has emerged as a key contributor of EV-based CRISPR delivery, offering key advantages such as transient RNP exposure, reduced off-target risk, and modular cargo compatibility. Nonetheless, shared challenges persist, including dependence on fusogenic viral proteins, limited in vivo editing efficiency, incomplete control over cargo release, and unresolved scalability and regulatory constraints, which collectively define the critical barriers to clinical translation.

EV-mediated delivery of mRNA via scaffold-protein engineering

Scaffold-protein engineering has emerged as a promising strategy to overcome the intrinsically low efficiency and poor controllability of passive mRNA loading into EVs. By genetically fusing EV-resident membrane proteins to RNA-binding modules, these approaches actively recruit defined mRNA cargoes during EV biogenesis, markedly increasing loading efficiency, intraluminal protection, and functional delivery.

Early proof-of-concept systems, such as the CD9-HuR platform, demonstrated that tethering RNA-binding proteins to TSPANs enables selective enrichment of motif-bearing RNAs and functional in vivo delivery, particularly to the liver [136]. Subsequent platforms introduced programmable RNA-protein interaction pairs, including MCP/MS2 and L7Ae/C-D box, allowing quantitative and sequence-defined mRNA loading. In familial hypercholesterolemia models, CD9-MCP-based systems successfully restored LDLR expression and reduced atherosclerotic burden [134]. However, these studies also revealed a recurring limitation: strong mRNA tethering can compromise cytosolic release and translation, necessitating additional “releaser” strategies or competitive displacement mechanisms.

More advanced designs address this retention-release trade-off by integrating multiple functions into a single scaffold. The PTGFRN-Δ687 combines mRNA loading, immunoaffinity purification of high-payload EV subpopulations, and protease- or acidification-triggered cargo release, achieving comparable therapeutic efficacy at substantially reduced doses in vivo [160]. Parallel efforts have expanded the scaffold repertoire beyond canonical TSPANs. A systematic screen identified truncated PlexinA1 (PLXNA1) as a highly efficient EV-sorting scaffold capable of robust mRNA and protein loading while supporting modular surface display [157].

Among current platforms, the CD63-PUFe TAMEL system represents a benchmark, achieving > 200-fold mRNA enrichment with approximately 89% intraluminal protection and potent antitumor efficacy at picogram-per-kilogram doses [120]. Related CD63-L7Ae/C-D box systems have enabled multiplexed therapeutic mRNA delivery for MI and CRISPR-Cas 12a-based HIV genome editing, demonstrating broad versatility across disease contexts [122, 123].

Despite these advances, key translational barriers remain. Endosomal escape is frequently inefficient and often augmented by viral fusogens such as VSV-G or Nipah virus F/G proteins, raising immunogenicity and safety concerns [120, 178]. In addition, EV heterogeneity, liver-biased biodistribution, genetic instability of producer cell lines, and scalable manufacturing continue to limit clinical readiness. Overall, scaffold-protein engineering has transformed EVs into programmable, high-potency mRNA delivery vehicles. Future progress will depend on non-viral endosomal escape solutions, next-generation scaffolds with balanced retention-release kinetics, and production strategies compatible with clinical translation.

Sources of scaffold protein-engineered EVs

HEK293T cells are widely used (70%, N = 149) as EV producer cells due to the rapid proliferation rates and the ability to be easily manipulated to enhance EV content or surface composition (Fig. 6). HEK293T-derived EVs have been applied to deliver chemotherapeutic agents and therapeutic protein constructs in a schwannoma model, as well as miRNA therapeutics for breast cancer [321]. These properties make HEK293T EVs particularly well suited for systematic scaffold protein screening, quantitative cargo loading evaluation, and mechanistic dissection of EV biogenesis and delivery. However, the intrinsic biological activity of EVs is determined by the parent cell and the environmental conditions from which they originate [322]. Therefore, HEK293T-derived EVs primarily function as a neutral platform for engineering purposes and provide limited insight into the behavior of engineered EVs within complex biological or disease-specific microenvironments.

Fig. 6.

Overview of EV sources and scaffold proteins used in scaffold-based EV engineering. (a) Distribution of donor cell types employed for scaffold protein-engineered EVs. EVs are predominantly derived from HEK293-related cell lines, accounting for approximately 70% of reported studies. Other donor sources include MSCs, DCs, and a variety of additional cell types such as HepG2, HeLa, B16-F10, Jurkat, C2C12, NK cells, and U87 cells, each contributing a relatively small proportion. (b) Heatmap summarizing the scaffold proteins utilized in EV engineering from different donor cell categories, including HEK293-related cells, tumor and other cell lines, stem cells and primary cells, immune cells, and in vivo sources. Scaffold proteins are involved in both cargo loading and surface display strategies. Color intensity indicates the number of publications reporting the use of each scaffold-cell type combination. Scaffold proteins reported in only a single study are excluded to emphasize commonly adopted engineering strategies. Prominent scaffolds include classical EV markers (e.g., CD63, CD9, CD81, Lamp2b), viral components (e.g., Gag, VSVG), and adaptor or lipid-anchoring domains (e.g., ARRDC1, C1C2, Myr). All quantitative data were derived from Supplementary Table 1

Compared with HEK293T-derived EVs, immune cell-derived EVs possess intrinsic immunological functions [323]. For example, DC-derived EVs carry key molecules required for the activation of anti-tumor T cell-mediated immune responses and can interact with T lymphocytes, thereby inducing anti-tumor immune responses [324]. In addition, DC-EVs also can serve as effective mucosal adjuvants to enhance the immunogenicity of influenza HA vaccines [325]. Immune cell-derived EVs retain their intrinsic immunological activity and can be further engineered to achieve more precise modulation of immune responses. For instance, DC-EVs were genetically engineered by fusing a CAP to the N-terminus of Lamp2b, enabling surface display of CAP on EVs. These engineered EVs were used to deliver CRISPR-Cas9 via intra-articular (IA) administration, achieving targeted knockdown of MMP-13 and thereby alleviating or preventing cartilage degeneration in a rat model of osteoarthritis [237]. Engineering NK cell-derived EVs with surface-displayed agonistic DR5 scFvs enables cooperative induction of DR5-mediated apoptosis and intrinsic cytotoxicity of NK cell-derived EVs, resulting in efficient elimination of DR5-positive cells [280].

Similarly, stem cell-derived EVs represent another major class of EVs with intrinsic regenerative and immunomodulatory properties. Stem cell-derived EVs, particularly MSC-EVs, have been widely reported to exert anti-inflammatory effects, promote tissue repair and regeneration, enhance angiogenesis, and inhibit apoptosis across diverse disease models [326, 327]. For instance, by displaying a CAP on the EV surface via Lamp2b fusion and encapsulating miR-199a-3p within the vesicles, the engineered MSC-derived EVs enabled efficient delivery of miR-199a-3p into chondrocytes, penetrated deep articular tissues in vivo, and exerted pronounced protective effects on damaged cartilage in a DMM-induced osteoarthritis mouse model [238]. In addition, MSC-EVs were engineered to display GPI-anchored insulin on the surface for enhanced cellular targeting, while Sirtuin3 was enriched within the vesicles as an internal cargo to restore mitochondrial function in myocardial ischemia-reperfusion injury [299].

In summary, EVs derived from different parent cell types exhibit distinct intrinsic properties that determine their functional roles and engineering suitability. HEK293T-derived EVs serve as a neutral and highly controllable platform for mechanistic and screening studies, whereas immune cell- and stem cell-derived EVs retain cell type-specific biological activities that can be further harnessed for immunomodulatory and regenerative applications. Accordingly, the choice of EV producer cell type should be considered by the intended application, requiring a balance between engineering flexibility and intrinsic biological functionality. Scaffold protein usage shows broadly overlapping patterns across EV sources, with modest source-associated preferences, where HEK293-derived EVs support greater scaffold diversity for method development, while stem cell- and immune cell-derived EVs converge on a limited set of Lamp2b- and tetraspanin-based scaffolds.

Manufacturing and analytical considerations for scaffold protein-engineered EVs

The successful clinical translation of scaffold protein-engineered EVs requires a fundamental transition from laboratory-scale production to robust, scalable, and Good Manufacturing Practice (GMP)-compliant manufacturing. In this context, Chemistry, Manufacturing, and Controls (CMC) considerations are not auxiliary but central determinants of product safety, consistency, and therapeutic performance [73]. EVs are complex and heterogeneous biologics whose identity and quality are intrinsically linked to the source cell, engineering strategy, and downstream processing (Fig. 7). Consequently, a well-defined CMC framework is indispensable for regulatory acceptance and clinical reproducibility [328].

Fig. 7.

Manufacturing process for scaffold protein-engineered EVs. Schematic overview of a representative end-to-end manufacturing process for scaffold protein-engineered EVs, spanning upstream cell engineering to downstream purification and quality control. The upstream phase includes genetic modification of producer cells to express scaffold-cargo or scaffold-targeting fusion constructs, with key considerations such as transfection efficiency, genetic stability, cell viability, proliferation, and induced functional alterations. EV production is subsequently carried out using scalable culture systems (e.g., planar or bioreactor-based formats), where maintenance of cell phenotype stability, controlled culture conditions, process scalability, and batch-to-batch consistency are critical. Downstream processing involves EV isolation and purification using methods such as tangential flow filtration (TFF), chromatography, or ultracentrifugation, with emphasis on yield, recovery efficiency, process robustness, and manufacturing cost. Final quality control encompasses comprehensive characterization of EV size, concentration, purity, marker expression, cargo loading, and functional potency to ensure product consistency and suitability for translational or clinical applications

Upstream manufacturing: cell banks, bioreactors, and EV production

Upstream manufacturing establishes the foundation for EV quality and scalability. Clinical-grade EV production typically begins with well-characterized master and working cell banks derived from defined producer cell lines, such as HEK293 or mesenchymal stromal cells, cultured under serum-free and xeno-free conditions [329, 330]. The lipid and protein composition of EVs critically influences their pharmacokinetic properties, and their native constituents may contribute to enhanced bioavailability and reduced adverse effects [61]. Therefore, for scaffold protein-engineered EVs, genetic modification of producer cells must be tightly controlled to ensure stable expression of scaffold constructs without compromising EV biogenesis, composition, or biological activity.

To enable large-scale production, bioreactor-based culture systems—including stirred-tank, hollow-fiber, or fixed-bed bioreactors—have increasingly replaced planar flask-based cultures [331]. Bioreactor-based culture systems, such as hollow fiber bioreactors, enable high-density cell culture with continuous nutrient exchange, enabling sustained and large-scale production of EVs and reducing batch-to-batch variability [332]. Importantly, these upstream platforms are designed to be compatible with closed and scalable downstream processing workflows, facilitating the transition from EV production to GMP-compliant purification and formulation [333].

However, several challenges remain. Many clinically advanced platforms still rely on adherent HEK293T cultures grown in planar or multilayer flasks, which limits scalability for commercial manufacturing [334]. In addition, dependence on specific producer cell lines and the inherent heterogeneity of EV populations complicate standardization across batches and facilities [335].

Downstream processing: isolation and purification technologies

Downstream processing exerts a profound influence on EV yield, purity, scalability, and batch consistency. Each isolation method involves inherent trade-offs, and no single technology currently satisfies all manufacturing requirements (Table 3).

Table 3.

Comparison of purification technologies for clinical-grade EVs

Method	Yield (Recovery)	Purity	Scalability	Cost	Batch Variability
Differential Ultracentrifugation (dUC) / Density Gradient	Low to Medium	High (with gradient)	Low	Low (Equipment high)	High
TFF	High	Medium	High	Medium	Low
Size-Exclusion Chromatography (SEC)	Medium	High	Low to Medium	High	Low
Immunoaffinity	Low	Very High	Low	Very High	Medium

dUC, historically the most widely used laboratory method, suffers from limited scalability, operator-dependent variability, and co-isolation of protein aggregates and non-vesicular particles [336, 337]. Density gradient centrifugation offers superior purity and effective separation from lipoproteins but is labor-intensive and poorly suited for large-scale or GMP manufacturing [328].

TFF has emerged as a cornerstone technology for clinical-scale EV manufacturing. TFF enables closed-system processing of large volumes, efficient concentration, and buffer exchange, making it well aligned with GMP requirements [338, 339]. Nevertheless, TFF alone does not fully discriminate EVs from similarly sized non-vesicular extracellular particles, necessitating additional downstream polishing steps to achieve higher purity [340].

SEC is therefore widely employed as a complementary purification strategy due to its ability to preserve EV integrity while effectively removing soluble protein contaminants. When integrated with upstream bioreactor-based production, combined TFF-SEC workflows have been successfully implemented for large-scale purification of EV-rich conditioned medium under cGMP-compatible conditions. Such integrated processes significantly reduce macromolecular protein impurities while maintaining EV size, morphology, and bioactive surface display, demonstrating that established bioprocessing unit operations can be adapted for scalable purification of engineered EVs [333]. However, the limited loading capacity of SEC columns and the inherent sample dilution during elution continue to constrain throughput and scalability [337, 341, 342].

Immunoaffinity-based capture strategies offer exceptional specificity by isolating defined EV subpopulations based on surface markers, but their high cost, limited binding capacity, and challenges in gentle vesicle elution restrict their applicability for large-scale therapeutic manufacturing [328].

In practice, hybrid downstream workflows—most commonly combining TFF for volume reduction and buffer exchange with SEC or filtration-based polishing—are increasingly favored to balance recovery, purity, scalability, and regulatory compliance in the manufacturing of scaffold protein-engineered EVs [337, 343, 344].

Critical quality attributes (CQAs), analytical characterization, and release criteria

The definition and control of Critical Quality Attributes (CQAs) form the cornerstone of EV CMC development. Current regulatory thinking, together with the MISEV2023 guidelines, emphasizes a multiparametric characterization strategy rather than reliance on any single analytical readout (Table 4) [345].

Table 4.

Critical quality attributes & analytical methods for clinical-grade EVs

CQA Category	Attribute	Analytical Method	Typical Acceptance Range	Reference
Appearance	Color, Clarity, etc.	Visual Inspection	Clear to opalescent, no visible particulates	[343]
Physical	Size / PDI	NTA	50–200 nm; PDI < 0.3	[345, 346]
Chemical	pH	Potentiometry	7–8	[347]
Surface charge	Zeta potential	DLS	−10 to − 30 mV	[328]
Concentration	Particle Concentration	NTA / Fluorescent NTA	> 1 × 109 particles/mL	[343]
	Total Protein Concentration	BCA Assay	Report value (e.g., < 1 mg/mL)	[343]
Identity	CD9/CD63/CD81	Western blotting, FC, ELISA	Positive	[343]
Cytosolic markers	TSG101 / Alix	Western blotting	Positive	[345]
Negative markers	Calnexin, TOMM20	Western blotting	Undetectable	[348]
Scaffold	Scaffold protein level	Western blotting / MS	Batch-consistent	[349]
Cargo	RNA/protein payload	Western blotting / qPCR / HPLC	Defined range	[349]
Safety	Endotoxin	LAL	< 50 EU/µL	[328]
	Sterility	Direct Inoculation / BacT/ALERT	No growth (14 days)	[328, 343]
	Mycoplasma	qPCR	Negative / <10 CFU/mL	[343]
Residual DNA/RNA	PicoGreen	< regulatory limit		[343]
Potency	MoA-specific assay	Functional assay	Pass/fail	[348]
Stability	Freeze-thaw	NTA, potency	≤ X% change	[347]

Physical attributes, including particle size, size distribution, and surface charge, represent primary CQAs. Small EVs typically exhibit modal diameters between 50 and 200 nm [345]. Nanoparticle tracking analysis (NTA) remains the most widely applied method for assessing particle size and concentration; however, it lacks specificity for vesicular particles and may overestimate size relative to cryo-electron microscopy [328, 348]. The polydispersity index (PDI) serves as an indicator of batch homogeneity, while zeta potential—generally negative due to lipid membrane composition—affects colloidal stability, biodistribution, and cellular uptake [73, 346].

Identity confirmation relies on enrichment of canonical EV markers, including the TSPANs CD9, CD63, and CD81, as well as cytosolic membrane-associated proteins such as TSG101 and Alix [343, 345]. Purity is further supported by depletion of intracellular markers, including calnexin or TOMM20, indicating minimal contamination from cellular debris or organelles [348].

For scaffold protein-engineered EVs, quantitation of both the scaffold and therapeutic cargo constitutes a particularly critical CQA. Key parameters include cargo loading efficiency, cargo-to-particle or cargo-to-protein ratios, and batch-to-batch consistency [349]. Early clinical studies, such as those involving siRNA-loaded EVs, employed total EV protein content and defined stoichiometry between vesicles and nucleic acid payloads as pragmatic release metrics, highlighting the need for mechanism-informed yet operationally feasible quantitation strategies [350].

Safety-related impurities represent another essential category of CQAs. These include non-vesicular extracellular particles, lipoproteins, residual host cell proteins, endotoxin, and host cell DNA, all of which must be tightly controlled to mitigate immunogenicity and oncogenic risk [343, 348].

Functional potency is a central CQA and an essential component of clinical lot release. Potency assays should reflect the intended mechanism of action and demonstrate consistent biological activity across batches [73]. Depending on therapeutic application, such assays may include macrophage polarization for immunomodulatory EVs, endothelial proliferation or tube formation for regenerative indications, or target gene knockdown for RNA-based therapeutics [343, 348, 350]. Although no universal potency assay exists for EV products, functional relevance and reproducibility are key regulatory expectations.

Stability represents a lifecycle CQA that directly impacts storage, distribution, and clinical usability. Most EV formulations exhibit optimal stability at − 80 °C, with reported shelf lives extending to several years; however, repeated freeze-thaw cycles may compromise vesicle integrity and cargo activity [347, 351]. Emerging strategies such as lyophilization with cryoprotectants (e.g., trehalose) offer opportunities for improved storage and distribution but require careful optimization to preserve EV structure and biological function [349, 352].

Outlook of manufacturing for scaffold protein-engineered EVs

As scaffold protein-engineered EVs progress toward clinical application, CMC considerations will increasingly define translational success. Harmonization of upstream manufacturing platforms, scalable downstream purification workflows, and standardized analytical frameworks for CQAs and potency testing are urgently needed. Continued coordination among academia, industry, and regulatory agencies will be essential to transform engineered EVs from experimental platforms into reproducible and clinically viable biologics.

Discussion and future perspectives

In recent years, the engineering of EVs based on scaffold proteins has advanced rapidly. For example, research has identified dozens of scaffold proteins that can be used for the engineering of EVs, with over a hundred relevant research reports, and a few engineered EVs based on scaffold proteins have entered the clinical trial stage (Table 1). However, this field still faces some unresolved issues. There are some directions for future improvement.

In terms of biological mechanisms, we need to conduct in-depth analysis of the structure-function relationship of scaffold proteins. During the engineering process, we need to further clarify whether and how the overexpression of scaffold proteins will affect the production, composition, and properties of EVs. In addition, analysis based on structure and function may identify key fragments from scaffold proteins for better EV engineering. Short scaffolds can improve the protein expression level in genetic engineering and enhance the loading efficiency. We also need to compare and analyze the differences in the effects of different scaffold proteins in loading different cargos (including proteins, nucleic acids, genome editing tools, and even viral vectors such as AAV). This will guide us in selecting different scaffold proteins for different cargos. Moreover, it is still unclear whether EVs from different cell sources require different scaffolds. Currently, EV engineering based on scaffold proteins mainly focuses on the tool cell HEK293 (Fig. 6), and there is relatively little research on the engineering of EVs from functional cells such as stem cells and immune cells.

Currently, there have been many studies on using scaffold proteins to load cargos, but how to load and release cargos in a controllable manner remains a problem. Some cargos need to be released after being delivered to target cells to enter specific subcellular organelles. For example, gene editors need to enter the nucleus to function. Although preliminary studies have shown that sequences such as intein can achieve cargo release, it is necessary to do more research to prove and improve its efficiency and robustness. Developing more efficient strategies for switch-type controllable loading and release will be able to promote the application of engineered EVs. Additionally, reprogramming the internal physicochemical environment of EVs through genetic engineering can improve the encapsulation efficiency and stability of sensitive cargos. For instance, mRNA is highly negatively charged [353], while the internal environment of EVs is complexed due to varieties of cargos inside and their inner membrane surface may be negatively charged because the inner side of the cell membrane and the outer surface of the intracellular membrane system are negatively charged [354–356], so the loading efficiency of EVs for mRNA is usually low. Modifying the internal charge environment of EVs based on genetic engineering schemes may improve the loading efficiency. Currently, most studies use two different scaffold proteins for loading and targeting modification respectively [121, 224]. Therefore, another promising approach is to develop more multifunctional scaffolds that can simultaneously mediate internal cargo loading and external targeting, which can simplify EV engineering and improve efficiency. On the other hand, in the more distant future, we may use synthetic biology to implement multi-pathway parallel engineering schemes based on scaffold proteins to obtain more complex customized engineered EVs. Meanwhile, recent advances in artificial intelligence (AI) further accelerate EV engineering. Machine learning (ML) is the key tool to unlock the potential of EV analysis. ML-based EV analysis has shown strong potential in cancer diagnosis, including early detection, subtype classification, stage differentiation, and molecular profiling [357]. In parallel, AI provides powerful predictive capabilities to guide the rational design of engineered EVs and achieve precision control of drug delivery [358]. At present, the application of AI and ML in EV research is largely confined to EV characterization and heterogeneity analysis [359], rather than direct optimization of scaffold protein-based EV engineering. The direct use of AI for rational scaffold protein design, cargo-loading optimization, or controllable release in engineered EVs remains largely unexplored. However, advances in data-driven protein design [360] and nanoparticle biodistribution modeling [361] suggest that such approaches may become feasible as standardized EV datasets and quantitative benchmarks emerge [362].

Although significant progress has been made in the engineering of EVs based on scaffold proteins, there are still very few cases that have reached the clinical stage. The large-scale production of engineered EVs remains one of the primary challenges in their clinical application.

Despite their potential as versatile drug delivery systems, EVs are naturally secreted by cells at low quantities, limiting the scalability of their production [45, 363]. Current methods, such as cell culture-based production and isolation techniques, often face challenges related to consistency, reproducibility, and high yields [73, 329, 364]. Additionally, the isolation of EVs from cell culture supernatants or biological fluids can be cumbersome and time-consuming, requiring advanced purification techniques such as ultracentrifugation, SEC, and TFF [340, 365, 366]. These methods, while effective, are not always scalable and can be expensive [331]. Therefore, finding more efficient, cost-effective, and scalable methods for EV production is essential for the successful translation of engineered EVs into clinical therapies. Ongoing research is focusing on optimizing cell culture conditions, improving EV isolation methods, and exploring the use of bioreactors to enhance the quantity and quality of EVs produced [47]. In the future, more stable and reliable large scale purification schemes for EVs need to be developed. Engineered scaffold proteins can be fused with affinity tags or peptide motifs on the EV surface, thereby enabling convenient affinity purification in addition to functional modification. Establishing a stable cell line capable of secreting engineered EVs will also promote their application development.

Engineered EVs obtained through genetic engineering still face challenges such as quality control and safety evaluation before entering the clinic. Comprehensive preclinical safety evaluations of engineered EVs should be carried out to evaluate potential safety risks associated with various engineered EVs compared to natural EVs. For instance, some scaffold proteins, which play key roles in immune responses, may trigger immune reactions when modified, potentially affecting the safety and effectiveness of their clinical applications [367]. Addressing these challenges is crucial for fully realizing the therapeutic potential of EV engineering based on scaffold proteins and accelerating its transition from the laboratory to clinical applications.

Abbreviations

AAV

Adeno-associated virus

ABD

Albumin-binding domain

ABP

Aptamer-aptamer-binding protein

ACE2

Angiotensin-converting enzyme 2

acircRNAs

Artificial circular non-coding RNAs

AI

Artificial intelligence

ALS

Amyotrophic lateral sclerosis

Ang2

Angiopep-2

AR

Androgen receptor

AREs

AU-rich elements

ARMMs

ARRDC1-mediated Evs

ARRDC1

Arrestin domain-containing protein 1

BACE1

β-site amyloid precursor protein-cleaving enzyme 1

BASP1

brain acid-soluble protein 1

BBB

Blood-brain barrier

BDNF

brain-derived neurotrophic factor

BLyS

B lymphocyte stimulator

CAP

chondrocyte-affinity peptide

CBD

collagen-binding domain

CCR8

(C-C motif) receptor 8

CDC

cardiosphere-derived cells

cGAS

cyclic GMP-AMP synthase

CIBN

CRY2 blue light dimerization

CMC

Manufacturing and Control

CMP

Cardiac-specific peptides

CNS

Central nervous system

Col-I

collagen I

CQAs

Critical Quality Attributes

CRISPR

clustered regularly interspaced palindromic repeats

CRY2

Cryptochrome 2

CSDS

chronic social defeat stress

CTP

Cardiac-targeting peptide

DCs

Dendritic cells

DR5

Death receptor 5

dUC

differential ultracentrifugation

EGFR

Epidermal growth factor receptor

EHS

Exertional heat stroke

ESCRT-I

Endosomal sorting complex required for transport-I

eVLPs

Engineered virus-like particles

EVs

Extracellular vesicles

FKBP12

FK506-binding protein

FRB

FKBP12-rapamycin binding

Gag

Group-specific antigen

GBA

β-glucocerebrosidase

GBM

glioblastoma

GMP

Good Manufacturing Practice

GPI

Glycosylphosphatidylinositol

HA

Hemagglutinin

HBV

Hepatitis B virus

anti-HLA-G

human leukocyte antigen-G-VHH antibody-modified

HSV1

Herpes simplex virus 1

Ig

Immunoglobulin

IL

interleukin

IV

intravenous

Lamp2

lysosome-associated membrane protein 2

LDLR

Low-Density Lipoprotein Receptor

LEL

Large extracellular loop

LNP

Lipid nanoparticle

LPS

Lipopolysaccharide

MA

Matrix

MARCKS

Myristoylated alanine-rich C kinase substrate

MCP

MS2 bacteriophage coat protein

MFG-E8

Milk fat globule-epidermal growth factor-factor 8

MI

myocardial infarction

minRISC

minimal RNA-induced silencing complex

ML

Machine learning

MOR

Opioid receptor mu

MSC

Mesenchymal stem cells

MVBs

Multivesicular bodies

NanoMEDIC

Macromolecular cargo

NLS

Nuclear localization signal

NLuc

NanoLuciferase

N-Myr

N-myristoylation

NTA

Nanoparticle tracking analysis

OVA

Ovalbumin

PCa

Prostate cancer

PCSK9

Proprotein convertase subtilisin-kexin type 9

PDGFR

Platelet-derived growth factor receptor

PDI

Polydispersity index

PE-eVLP

Prime editor engineered virus-like particle

PEG

Polyethylene glycol

PhoCl

Photocleavable

PLXNA1

Plexin-A1

POIs

Proteins of interest

PS

Phosphatidylserine

PTGFR

Prostaglandin F2 receptor

PTGFRN

PTGFR inhibitor

PTTG1IP

pituitary tumor-transforming 1 interacting protein

RBD

receptor-binding domain

RNP

Ribonucleoprotein

RVG

Rabies virus glycoprotein

scFv

Single-chain variable fragment

SEC

Size-exclusion chromatography

SIRPα

Signal regulatory proteinα

SLE

systemic lupus erythematosus

sLeX

sialyl Lewis X

TFF

Tangential flow filtration

Tluc

ThermoLuc

TM3

transmembrane helix 3

TMD

Transmembrane domain

TSG101

Tumor susceptibility gene 101

UUO

Unilateral ureteral obstruction

VFIC

VSV-G-Foldon-Intein-Cargo

VSV-G

Vesicular stomatitis virus glycoprotein

Author contributions

Conceptualization: Qiang Wu. Writing Original Draft: Chaofan Zhang, Yue Wu. Literature Review: Chaofan Zhang, Yue Wu, Yuezhou Wang, Cunbo Yao, Mengting Ma, and Jiacong Li. Writing Review & Editing: Chaofan Zhang, Yue Wu, and Qiang Wu. Visualization: Yue Wu, Yuezhou Wang, and Cunbo Yao. Supervision: Qiang Wu. Project Administration: Qiang Wu. Funding Acquisition: Qiang Wu. All authors have read and approved the article.

Funding

This work was supported by grants from the National Natural Science Foundation of China (82400330), Basic Research Program of Jiangsu (SBK2024042497), the Natural Science Foundation of the Jiangsu Higher Education Institutions of China (24KJB180023) and Research Development Fund of Xi’an Jiaotong-Liverpool University (RDF-23-02-032).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Competing interests

The authors declare no competing interests.