Engineered small extracellular vesicles in hematologic malignancies: mechanisms, therapeutic strategies, and translational challenges

Abstract

Hematologic malignancies remain among the most challenging cancers to treat due to genetic heterogeneity, clonal evolution, and therapy resistance. Extracellular vesicles (EVs), particularly small EV (sEV)-enriched populations, have emerged as active mediators of disease biology, contributing to tumor progression, immune evasion, and chemoresistance through intercellular transfer of bioactive cargo. Recent advances in EV engineering have repositioned these vesicles as programmable delivery platforms capable of transporting nucleic acids, proteins, and chemotherapeutic agents with improved targeting potential. Preclinical studies across multiple hematologic models demonstrate that engineered EVs can induce immune activation, modulate oncogenic signaling pathways, and partially overcome drug resistance. However, these findings remain largely confined to experimental settings, with limited standardization of loading efficiency, biodistribution, and functional potency. Clinically, EV-based applications in hematology are still at an early stage, with most studies focused on biomarker discovery and supportive therapies rather than direct antitumor interventions. In parallel, theranostic EV platforms and liquid biopsy approaches offer promising opportunities for minimally invasive disease monitoring, although their clinical validation remains incomplete. Artificial intelligence (AI) further enhances this field by enabling advanced biomarker analysis and guiding cargo design and targeting strategies, yet its therapeutic applications are still largely exploratory. Despite key challenges, including vesicle heterogeneity, donor variability, suboptimal cargo loading, and manufacturing constraints, these limitations are primarily technical and may be addressed through standardization and engineering optimization. Collectively, EV-based systems represent a promising but still maturing platform with the potential to contribute to next-generation precision oncology in hematologic malignancies.

Introduction

Hematologic malignancies, including leukemias, myelodysplastic syndromes (MDS), lymphomas, and plasma cell neoplasms, constitute a diverse spectrum of clonal disorders of hematopoietic and lymphoid origin [1, 2]. These cancers remain a significant global health burden, with outcomes that vary widely due to differences in genetic architecture, clonal evolution, and therapeutic resistance [3, 4]. While conventional treatment strategies have relied on broad cytogenetic markers and clinical presentation, there is growing evidence that a deeper molecular understanding is essential to guide precision medicine approaches in hematology [5, 6].

Within this evolving landscape, extracellular vesicles (EVs), a heterogeneous population of lipid bilayer–delimited particles released by cells, have emerged as active biological mediators rather than passive byproducts [7, 8]. Small EVs (sEVs), often operationally defined based on size (< 200 nm), represent a commonly studied EV-enriched fraction; however, they may include vesicles of distinct biogenesis, including but not limited to exosomes [8]. According to MISEV2023 guidelines, the term “exosome” should be reserved for vesicles of endosomal origin, which is difficult to definitively establish in most studies; therefore, we use “EV” as the general term throughout this review and “sEV” as a size-based operational subset.

Initially considered cellular waste, EVs are now recognized as mediators of intercellular communication that modulate the tumor microenvironment, regulate immune surveillance, and contribute to drug resistance [9, 10]. Their ability to transfer proteins, nucleic acids, and lipids between cells provides a unique window into the molecular circuitry of hematologic malignancies, while also offering new therapeutic and diagnostic possibilities.

Recent advances in EV engineering have extended their role from natural conveyors of biological signals to promising programmable delivery systems, capable of transporting small molecules, nucleic acids, and even CRISPR-based gene-editing tools [11, 12]. Such engineered vesicles may help overcome drug resistance and improve targeting specificity in blood cancers [13]. In parallel, the development of theranostic EVs, vesicles designed to function simultaneously as biomarkers and therapeutic vehicles, underscores their potential in personalized oncology [14]. These constructs not only enable real-time disease monitoring via liquid biopsy but also provide minimally invasive platforms for targeted intervention.

Complementing these biological and technological advances, artificial intelligence (AI) and computational biology are increasingly integrated into EV research. AI-driven analytics can synthesize multi-omics data from EVs, refine patient stratification, and support optimization of engineering strategies for safety and efficacy [15–17]. Together, these converging innovations herald a paradigm shift in which EV biology, advanced bioengineering, and AI converge to redefine the diagnosis, monitoring, and treatment of hematologic malignancies.

This review seeks to address a central guiding question: how can engineered and theranostic EVs, together with AI-driven analysis, be translated into clinically actionable tools for hematologic malignancies? To address this, EV pathobiology is first examined to understand how these vesicles function as active disease modulators. Next, engineered EVs are discussed as platforms for targeted therapy, enabling precision delivery of drugs, RNA-based therapeutics, and gene-editing tools. Finally, theranostic EVs are considered as dual-purpose systems for disease monitoring and therapeutic intervention. In this review, we intentionally adopt a broad EV-centered framework to integrate findings across heterogeneous vesicle populations. While many studies historically use the term “exosome,” we reinterpret these findings within the EV/sEV framework in accordance with MISEV2023 to improve conceptual clarity and translational consistency. This focus on sEV-enriched fractions reflects their relative abundance, stability, and frequent use in both experimental and translational studies.

Method

This study was conducted as a narrative review with a structured literature search strategy. A search was performed across multiple databases, including PubMed, Scopus, Web of Science, and Google Scholar, to identify relevant studies on extracellular vesicles in hematologic malignancies.

The search strategy incorporated combinations of the following keywords: “extracellular vesicles,” “small extracellular vesicles,” “exosomes,” “hematologic malignancies,” “leukemia,” “lymphoma,” “multiple myeloma,” “engineered EVs,” “theranostics,” and “artificial intelligence.”

Studies published up to March 1, 2026 were considered. Articles were selected based on relevance to EV biology, engineering strategies, and translational or clinical applications in hematology. Both experimental and clinical studies were included, while non-English articles were excluded.

Although this review is not systematic, efforts were made to ensure comprehensive coverage, methodological transparency, and inclusion of both foundational and recent advances in the field.

Extracellular vesicles in hematologic malignancies: pathobiology and clinical implications

sEVs, previously referred to as exosomes, are nanoscale particles secreted by most cell types and characterized by complex molecular cargo [8] (see Table 1 for MISEV-based markers). Consistent with MISEV2023 recommendations, the term “sEV” is used here as an operational size-based definition rather than a strict biogenesis classification.

Table 1.

MISEV-aligned characterization framework for extracellular vesicles (EVs)

Class	Example markers	Purpose	Notes
EV-enriched markers	CD9, CD63, CD81, Syntenin-1	EV identity confirmation	At least 2 transmembrane markers plus 1 cytosolic protein are generally recommended
Cytosolic components	ALIX, TSG101, HSP70	Support EV identity at the protein level	Useful for confirming EV-associated intracellular cargo, but not for assigning a specific EV subtype
Negative controls (contaminants)	ApoA1, Albumin, GM130, Calnexin	Exclude non-EV particles (NVEPs)	Essential for purity assessment
Particle characterization	NTA / TRPS	Size distribution & concentration	Operational, method-dependent
Morphology validation	TEM / AFM / Cryo-EM	Structural confirmation	Representative imaging required
Source-specific markers (hematologic EVs)	CD19, CD20, CD38, CD138, CD33	Lineage tracing in hematologic malignancies	Recommended for disease-specific EV profiling

In hematologic malignancies, sEVs function as important mediators of disease progression by transferring oncogenic transcripts, regulatory RNAs, and functional proteins. They remodel the bone marrow microenvironment, promote immune evasion, and support clonal evolution [10, 18, 19]. Unlike soluble factors, sEVs provide a stable and protected transport system, enabling efficient intercellular communication and sustained propagation of malignant signals [10, 18, 19]. This dual role, both reflecting and shaping tumor biology, positions sEVs at the intersection of pathogenesis and therapeutic innovation.

sEVs drive tumor progression through both direct and indirect mechanisms. They enhance proliferation, survival, and invasion while reprogramming stromal and endothelial cells via factors such as PKM2, VEGF, and TGF-β [18, 20–24]. This includes metabolic reprogramming (e.g., GLUT1/MCT4 upregulation) and promotion of angiogenesis and niche adaptation through hypoxia-related signals [25–31]. In parallel, sEVs suppress immune responses by inhibiting T-cell, NK-cell, and dendritic cell functions, facilitating immune escape [32, 33].

sEVs also mediate chemoresistance by transferring resistance-associated miRNAs (e.g., miR-155, miR-21) and drug efflux proteins such as ABCB1/P-gp, thereby reducing intracellular drug accumulation and promoting survival [34–38]. This establishes a self-reinforcing cycle of resistance propagation, reframing therapeutic failure as a vesicle-mediated systems process. Targeting EV biogenesis or employing engineered decoy vesicles represents emerging strategies to disrupt this axis [39, 40].

The stability and accessibility of sEVs in biological fluids make them promising non-invasive biomarkers for hematologic malignancies [41]. Their molecular content reflects both genomic alterations and dynamic treatment responses, supporting their use in liquid biopsy applications [42].

EV-based biomarker applications can be considered within the framework of liquid biopsy, enabling minimally invasive detection and monitoring of hematologic malignancies [43]. However, it is important to distinguish between emerging and clinically validated applications, as most EV-based liquid biopsy approaches remain at the exploratory or early clinical stage, with limited validation in large prospective cohorts [44].

Beyond diagnostics, engineered sEVs are increasingly explored as theranostic platforms capable of simultaneously monitoring disease and delivering targeted therapy [17]. Integration with artificial intelligence further enhances this potential, enabling analysis of complex EV profiles, patient stratification, and optimization of therapeutic design with improved predictive capacity [7, 45, 46].

However, clinical translation remains partly limited by challenges including EV heterogeneity, variability in isolation methods, and lack of standardized potency assays [47]. Addressing these issues is essential to advance EV-based strategies toward clinically robust applications.

Engineered EVs: precision strategies for therapeutic delivery

EVs, particularly sEVs, have emerged as promising platforms for therapeutic engineering due to their intrinsic biocompatibility, stability, and ability to transfer bioactive cargo [48]. Nevertheless, native vesicles present challenges, including low yield, cargo variability, and potential instability in biological activity [49]. To address these limitations, engineering strategies have been developed and are broadly categorized into pre-isolation (cell-based) and post-isolation (vesicle-based) approaches [10].

In pre-isolation strategies, donor cells are genetically or chemically modified to produce EVs enriched with specific cargos or surface ligands, enabling controlled and biologically integrated cargo loading [50]. In contrast, post-isolation engineering involves direct manipulation of purified EVs using physical (e.g., electroporation, sonication), chemical (e.g., click chemistry), or hybrid approaches to enhance cargo loading and targeting efficiency [51].

Common loading techniques such as electroporation, sonication, incubation, and freeze–thaw cycles have been widely applied, although their efficiency and impact on vesicle integrity vary significantly [52]. For example, electroporation enables loading of nucleic acids into EVs but may result in low encapsulation efficiency (reported as < 0.05% in some studies) and cargo aggregation, potentially leading to overestimation of loading efficiency [53]. Additionally, electroporation can alter EV physicochemical properties, including particle size, surface charge, and protein composition, thereby affecting their biological function and stability [54]. In contrast, sonication-based loading methods can achieve higher drug encapsulation efficiencies (e.g., ~ 40% in some studies) by temporarily disrupting membrane integrity, although this may compromise vesicle structure and downstream uptake efficiency [55].

Surface engineering approaches, including click chemistry, ligand conjugation, and PEGylation, have been developed to improve targeting specificity, circulation time, and immune evasion, further expanding the therapeutic versatility of EVs [56, 57]. Despite these advances, major translational barriers persist, including lack of standardized protocols, variability in EV characterization, and limited reproducibility across studies, which collectively hinder regulatory approval and clinical implementation [55].

A critical translational consideration is that native EVs, particularly those derived from tumor cells, may retain oncogenic or immunomodulatory cargo capable of promoting resistance, immune evasion, or microenvironmental remodeling. To mitigate this risk, current strategies increasingly prioritize non-tumorigenic sources (e.g., MSC-, immune cell-, or iPSC-derived EVs), together with cargo reprogramming, surface engineering, and purification workflows that reduce endogenous pathogenic signals. Thus, engineered EVs are better understood as controlled delivery systems in which therapeutic activity is designed to be primarily cargo-driven rather than dependent on native vesicle biology. A systematic overview of EV/sEV surface engineering approaches and cargo loading strategies, including their underlying mechanisms, advantages, and limitations, is provided in Tables 2 and 3, respectively.

Table 2.

Surface engineering strategies for EVs/sEVs: methods, advantages, limitations, and applications in targeted delivery

Method	Description	Advantages	Disadvantages	Applications	References
Click Chemistry	Covalent linking of targeting ligands to exosomal surfaces	Highly specific, rapid, non-destructive	Requires prior chemical modification	Adding antibodies or ligands (e.g., CD33)	[58]
PEGylation	Addition of PEG chains to protect exosomes	Extends circulation time	Reduces receptor-mediated uptake	Reducing macrophage clearance	[59]
Genetic Surface Display	Donor cell engineering (e.g., Lamp2b fusion proteins)	Stable ligand expression	Requires cell modification	Scalable targeting strategies	[12]
Ligand/Peptide Conjugation	Surface addition of peptides, aptamers, or antibodies	Target-specific delivery	Limited surface stability	Targeting AML cells	[60]
Electrostatic/Hydrophobic Adsorption	Surface adsorption through charge or hydrophobic interactions	Simple, fast	Unstable, easily washed off	Temporary in vitro applications	[61, 62]

Table 3.

Cargo loading strategies for EVs/sEVs: comparative overview of methods, efficiency considerations, and therapeutic applications

Method	Description	Advantages	Disadvantages	Suitable for	References
Electroporation	Uses an electrical pulse to introduce RNA/siRNA into exosomes	Efficient for nucleic acids	May cause aggregation or membrane disruption	miRNA, siRNA, ASO	[77, 78]
Sonication	Uses ultrasonic waves to permeabilize the membrane for drug entry	High loading efficiency	May alter exosome structure	Chemotherapeutics, siRNA	[79, 80]
Freeze–Thaw Cycles	Repeated freezing and thawing disrupt the membrane	Simple, fast	Potential loss of exosome integrity	Small molecules, mRNA	[81]
EXPLOR System	Light-inducible protein loading	Highly specific	complex system	protein delivery	[82]
Surfactant (e.g., Saponin)	Creates pores to allow drug entry	High loading efficiency	Potential toxicity to recipient cells	Protein and RNA delivery	[83, 84]
Chemical Transfection (e.g., Exo-Fect)	Uses lipid-based kits to introduce RNA/proteins	RNA/miRNA compatible	Costly, requires specialized reagents	Experimental gene therapy	[85]
EXPLOR System	Utilizes targeting motifs and functional protein fusion	High precision, specific targeting	Technically complex	Protein therapeutics	[86]

Disease-specific applications

Multiple myeloma (MM)

MM is characterized by pronounced genetic heterogeneity and strong dependence on the bone marrow microenvironment, which together limit the durability of current therapeutic strategies [63]. Although treatment modalities such as monoclonal antibodies, autologous stem cell transplantation, and CAR-T cell therapy have improved patient outcomes, complete and sustained remission remains uncommon, largely due to persistent disease complexity and microenvironment-driven resistance [64, 65].

Against this background, engineered sEVs have emerged as promising adjunctive platforms. sEVs generated via genetic modification of donor myeloma cells to express membrane-bound HSP70, combined with endogenous tumor antigen loading, enhanced dendritic cell maturation and induced CD4⁺ and CD8⁺ T-cell responses in vivo (30 µg/mouse), supporting activation of both CTL and NK-mediated immunity [66]. However, the absence of quantitative efficacy metrics and restriction to preclinical models limit translational interpretation [66].

In parallel, RNA-based engineering approaches have enabled sEV-mediated gene regulation in MM. Electroporation-mediated loading of siRNA targeting the SFRS8/CACYBP/β-catenin signaling axis has demonstrated significant tumor growth inhibition in both patient-derived xenograft (PDX) and SCID mouse models, along with suppression of osteoclast activity and disease progression [67]. Despite these promising functional outcomes, key engineering parameters such as loading efficiency, biodistribution, and off-target effects were not quantitatively defined, limiting the ability to assess delivery precision and scalability [67].

Furthermore, monocyte-derived sEVs engineered via surface expression of anti-BCMA and loaded with bortezomib through co-incubation (loading efficiency ~ 13%) demonstrated enhanced cellular uptake (11–16-fold increase) and improved antitumor efficacy compared to free drug, along with modulation of the bone marrow microenvironment [13]. However, relatively low loading efficiency and limited control over drug encapsulation highlight ongoing challenges in engineering optimization [13].

Collectively, these studies demonstrate the functional versatility of engineered sEVs in multiple myeloma. However, all evidence remains preclinical, and challenges related to reproducibility, scalable manufacturing, and regulatory standardization continue to limit clinical translation.

Acute myeloid leukemia (AML)

AML is a rapidly progressive hematologic malignancy characterized by immune evasion, high relapse rates, and limited long-term survival, with 5-year overall survival rates typically below 30%. These challenges highlight the urgent need for more effective and targeted therapeutic strategies [68]. Engineered sEV-based strategies have been explored to improve targeting specificity and therapeutic delivery.

One prominent strategy involves targeted drug delivery using ligand-engineered sEVs. For example, mesenchymal stem cell–derived sEVs engineered via CD63 fusion proteins displaying TPO-mimic peptides demonstrated high binding affinity to c-Mpl-expressing AML cells, with optimized constructs (e.g., CD63-mTPO3) showing enhanced receptor interaction and efficient endocytic uptake [69]. Upon loading with daunorubicin, these engineered sEVs (m3Exos@DNR) effectively eliminated c-Mpl⁺ AML cells in both in vitro and in vivo models, while maintaining manageable toxicity profiles. However, quantitative loading efficiency and long-term toxicity data were not fully reported, and variability in targeting across AML subtypes (c-Mpl expression ~ 50–65%) may limit broader applicability [69].

In parallel, advanced immuno-engineering approaches such as PRIME sEVs have been developed through genetic surface display of multiple functional proteins, including anti-CD3 and anti-CLL-1 antibodies combined with immune modulators (PD-1 and CD70), enabling simultaneous tumor targeting and T-cell activation [70]. These engineered sEVs demonstrated potent and selective cytotoxicity against AML cells, with EC50 values ranging from approximately 47 to 149 ng/mL in CLL-1⁺ cell lines, while showing minimal toxicity toward CLL-1⁻ cells. Furthermore, exosome yields of approximately 200 µg (~ 2 × 10¹⁰ particles) per 100 mL culture highlight potential scalability; however, the complexity of multi-component engineering and lack of in vivo efficacy standardization remain key translational challenges [70].

RNA-based and combinatorial delivery systems further expand the functional capabilities of engineered EVs in AML. For instance, red blood cell–derived EVs (REVs) co-loaded with doxorubicin and siIDO1 using sonication and membrane insertion strategies (e.g., P9-PEG-DSPE functionalization) demonstrated efficient cellular uptake (~ 10⁶ particles per cell) and dose-dependent cytotoxicity across multiple AML cell lines at drug concentrations ranging from 0.2 to 10 µg/m [71]. This platform induced immunogenic cell death (ICD) and IDO1 suppression, leading to enhanced dendritic cell maturation, T-cell activation, and cytotoxic T lymphocyte (CTL) recruitment in vivo, thereby integrating chemotherapy with immune activation. Nevertheless, variability in loading consistency, potential off-target immune activation, and lack of standardized pharmacokinetic profiling remain significant barriers to clinical translation [71].

Collectively, these studies demonstrate that the therapeutic efficacy of engineered sEVs in AML is highly dependent on the integration of targeting ligands, cargo selection, and immune modulation strategies. While ligand-directed delivery, multi-functional immuno-engineering, and chemo-immunological cascade systems each provide distinct advantages, the absence of standardized engineering protocols, incomplete quantitative characterization in some systems, and lack of clinical validation continue to limit their translational progression.

Other hematologic malignancies

In myeloproliferative neoplasms (MPN), engineered MSC-derived vesicles carrying USP5 inhibitors (USP5@EVs-CP) have been shown to selectively target JAK2V617F-mutated stromal cells and suppress aberrant proliferation, suggesting a niche-focused therapeutic strategy [72]. However, despite clear mechanistic targeting and in vivo efficacy, quantitative drug loading efficiency and pharmacokinetic profiles were not fully characterized, and the reliance on niche-specific targeting may limit applicability across heterogeneous disease states [72].

In chronic lymphocytic leukemia (CLL) and related B-cell malignancies, engineered sEVs have primarily focused on overcoming immune dysfunction [73]. EV-based approaches have therefore been explored using multiple strategies. Vesicles engineered to carry CD154 and EBV gp350 have been reported to enhance antigen presentation [74]. Nevertheless, this strategy is inherently dependent on EBV seropositivity and antigen-specific immunity, which may restrict its broader clinical applicability and introduce variability in patient responses [74].

More advanced engineering approaches have been applied in aggressive B-cell lymphomas. In mantle cell lymphoma (MCL), CAR-based strategies targeting B-cell antigens have demonstrated high response rates (e.g., overall response rates up to ~ 87% with targeted therapies in clinical settings), supporting the rationale for translating CAR functionality into vesicle-based systems [75]. Engineered vesicles delivering cytotoxic payloads such as doxorubicin via CAR-like targeting mechanisms have shown potent in vitro cytotoxicity, highlighting their potential as cell-free analogs of CAR-T therapy. However, these systems remain largely proof-of-concept, with limited data on biodistribution, in vivo persistence, and large-scale manufacturing feasibility [75]. Similarly, sEV co-loaded with rituximab and siPDK4 successfully restored CD20 expression and reversed rituximab resistance in diffuse large B-cell lymphoma, highlighting a new strategy for overcoming therapeutic resistance [76].

Collectively, these studies highlight that engineered sEVs can be tailored for niche targeting, immune reprogramming, or drug resistance reversal across diverse hematologic malignancies. However, despite strong mechanistic and preclinical efficacy, limitations in quantitative characterization, scalability, and patient-specific variability continue to represent major barriers to clinical implementation.

Advanced engineered ev platforms and sources

Chimeric antigen receptor-small extracellular vesicles (CAR-EVs)

CAR-derived EVs represent a specialized class of vesicles integrating antigen-specific recognition with nanoscale delivery. While CAR-T cell therapy has revolutionized hematologic malignancy treatment, it is associated with toxicities such as cytokine release syndrome and immune evasion [87–90].

CAR-sEV, derived from CAR-T cells, retain therapeutic efficacy while mitigating safety concerns [91]. These vesicles carry functional CAR molecules enabling antigen-specific targeting [92, 93], and their nanoscale size facilitates tissue penetration and traversal across biological barriers [94, 95]. Notably, they lack PD-1 expression, making them resistant to tumor-mediated immunosuppression, while carrying cytotoxic proteins such as perforin, granzymes, and Fas ligand [91].

Preclinical evidence has further underscored the versatility of this platform. Si et al. engineered EV to display anti-CD3/CD28 scFvs and to encapsulate CAR mRNAs using a LAMP-2B–MS2 system, thereby activating T cells and efficiently transferring CAR constructs [96]. Xu et al. subsequently demonstrated that anti-CD19 CAR-EVs loaded with CRISPR/Cas9 plasmids targeting MYC selectively entered CD19⁺ malignant cells and produced significant antitumor effects in vitro and in vivo [97]. Comparative analyses show that both CAR-T cells and CAR-EVs exert cytotoxicity; however, quantification of CAR protein content in EVs remains inconsistent, highlighting the need for standardized assays [98, 99].

CAR-engineered vesicles delivering doxorubicin demonstrated potent cytotoxicity in mantle cell lymphoma cells in vitro, but the scalability and biodistribution of such systems remain incompletely defined [75]. Similarly, EVs co-loaded with rituximab and siPDK4 successfully restored CD20 expression and reversed rituximab resistance in diffuse large B-cell lymphoma, highlighting a new strategy for overcoming therapeutic resistance [76].

Importantly, clinical studies have confirmed the persistence and functional activity of CAR-EVs in patients with B-cell malignancies. CD19.CAR⁺EVs were detected in peripheral blood as early as day + 1 post-infusion and remained measurable for up to two years, even after parental CAR T cells had disappeared. These vesicles retained CAR expression and cytotoxic molecules such as perforin and granzyme B, exerting antigen-specific killing of CD19⁺ target cells. Their long-lasting circulation suggests their utility as non-invasive biomarkers of CAR T-cell persistence and efficacy, while also supporting their potential as scalable, cell-free immunotherapeutics [100]. Moreover, they have shown potential as early biomarkers for immune effector cell–associated neurotoxicity syndrome (ICANS), outperforming traditional CAR-T expansion metrics [101].

Yet, critical challenges such as heterogeneity, variable CAR quantification, and the development of Good Manufacturing Practice (GMP)-compliant production strategies must be resolved to enable their clinical translation [102]. A comparative overview of the key features, advantages, and limitations of CAR-T cells versus CAR-EVs in hematologic malignancies is provided in Table 4.

Table 4.

Comparison of CAR-T cells and CAR-EVs: therapeutic mechanisms, functional properties, and safety profiles

Dimension	CAR-T Cells	CAR-EVs
Therapeutic nature	Live, genetically engineered T lymphocytes with CAR constructs integrated into genome	Cell-free vesicles secreted by CAR-T cells, naturally enriched with CAR proteins and effector molecules
Antigen recognition	Direct CAR binding at the immunological synapse	CAR displayed on vesicle surface mediates specific tumor antigen binding
Cytotoxic payload	Release of perforin, granzymes, FasL, cytokines upon cell–cell contact	Intrinsic cargo includes perforin, granzyme A/B, FasL; can be additionally loaded with RNAs/CRISPR or drugs
Persistence in vivo	Long-term engraftment possible, but risk of exhaustion, senescence, or deletion	Detected in circulation up to 2 years; resistant to PD-1/PD-L1 checkpoint inhibition
Tumor penetration	Limited infiltration in solid tumors due to cell size and TME barriers	Nanometer scale enables crossing blood–tumor and blood–brain barriers with improved tissue penetration
Safety profile	Associated with CRS, ICANS, off-target cytotoxicity, insertional mutagenesis	Lower risk of CRS; no genetic insertion risk; may serve as ICANS biomarker; potential neurotoxicity still under study
Biomarker utility	Expansion kinetics monitored by qPCR/flow cytometry	Circulating CAR⁺EV levels correlate with CAR-T activity and predict ICANS earlier than CAR-T expansion

MSC- and iPSC-derived EVs

Mesenchymal stem cell (MSC)-derived EVs exhibit potent immunomodulatory and regenerative properties, including suppression of T-cell activation and promotion of regulatory immune responses [103–106]. hey also contribute to tissue repair, angiogenesis, and hematopoietic niche remodeling [10]. However, their role is dual, as they may also promote tumor progression and drug resistance, necessitating careful evaluation in therapeutic contexts [10]. EVs generated from MSCs increasingly demonstrate the ability to replicate the immunomodulatory functions of MSCs, affecting B cells, T cells, NK cells, dendritic cells, and macrophages [107, 108].

Induced pluripotent stem (iPS) cells are pluripotent stem cells reprogrammed from adult somatic cells to an embryonic stem (ES)-like state via the enforced expression of genes and proteins that sustain ES cell characteristics [109–111]. Beyond EVs derived from iPSCs, iPSC-derived mesenchymal stem cells (iMSCs) have shown plasticity and immunomodulatory capabilities, offering enhanced survival, proliferation, and differentiation potential compared to adult MSCs [112]. Human iPSC-derived exosomes (hiPSC-Exos) have been shown to promote wound healing and nerve regeneration due to their low immunogenicity, while human MSC-derived exosomes (hMSC-Exos) exert strong anti-inflammatory and angiogenic effects, making them valuable in tissue repair and cancer suppression [113]. (Table 5)

Table 5.

Comparative characteristics of MSC-derived EVs and iPSC-derived EVs: source, biological functions, and therapeutic potential

Dimension	MSC-Exosomes (MSC-Exos)	iPSC-Exosomes (iPSC-Exos)
Cellular origin	Derived from adult mesenchymal stem cells (bone marrow, adipose tissue, umbilical cord)	Derived from reprogrammed induced pluripotent stem cells (iPSCs)
Immunogenicity	Generally low, but donor–recipient HLA mismatch may elicit mild immune responses	Very low; can be engineered into “universal” exosomes independent of MHC restriction
Biological functions	Immunomodulation, anti-inflammatory activity, angiogenesis, tissue repair, regulation of hematopoiesis	Delivery of RNAs, proteins, CRISPR constructs; pluripotency-associated regenerative effects; versatile therapeutic cargo loading
Applications in hematology	Mitigation of bone marrow fibrosis in myeloproliferative neoplasms; modulation of immune effector cells (B, T, NK, DCs, macrophages)	Potential in gene editing for inherited marrow failure, targeted therapy in leukemias and lymphomas, universal delivery platforms
Regenerative potential	Promotes fibroblast activity, collagen/elastin secretion, angiogenesis; reduces scar formation	Promotes wound healing and nerve regeneration; higher plasticity than adult MSC-Exo

CRISPR/Cas9-enabled EV platforms

The combination of CRISPR/Cas9 systems with sEV delivery represents a transformative approach in cancer therapeutics. Compared with viral or synthetic carriers, sEVs provide improved stability, greater targeting precision, and reduced risk of off-target editing [114]. Importantly, emerging preclinical evidence suggests that CRISPR-loaded EV platforms can be adapted for hematologic malignancies, although direct disease-specific data remain limited. For instance, in a preclinical study, CAR-engineered vesicles can deliver CRISPR/Cas9 components to B-cell malignancies with high efficiency, successfully editing the MYC oncogene without detectable off-target effects. These results underscore the potential of engineered vesicles to complement, and perhaps eventually surpass, CAR-T workflows in genome editing [97].

However, much of the mechanistic understanding of CRISPR-loaded EV delivery still derives from studies in solid tumor or non-hematologic models. For example, tumor-derived EVs have been shown to efficiently deliver CRISPR/Cas9 targeting PARP-1 in ovarian cancer models, resulting in gene disruption, induction of apoptosis, and enhanced chemosensitivity to cisplatin, driven in part by intrinsic tumor tropism of the vesicles [115]. Similarly, hybrid exosome–liposome nanoparticles have been developed to overcome the intrinsic limitation of EVs in packaging large nucleic acids, enabling efficient delivery of CRISPR/Cas9 plasmids into otherwise hard-to-transfect mesenchymal stem cells, thereby demonstrating a scalable strategy for genome-editing cargo loading [116].

Although these studies are not performed in hematologic malignancies, they provide critical proof-of-concept evidence for two key translational principles directly relevant to blood cancers: (i) EV-mediated cell-type–selective targeting and (ii) improved loading efficiency for large CRISPR constructs. Given that hematologic malignancies are often driven by well-defined genetic alterations (e.g., MYC dysregulation, BCL2 overexpression, or FLT3 mutations), these delivery platforms could be adapted to target disease-specific oncogenic drivers in a more precise manner.

Nevertheless, it is important to emphasize that current evidence supporting CRISPR-loaded EVs in hematologic malignancies remains largely preclinical and indirect. No robust clinical data are currently available, and several translational barriers persist, including efficient large-scale production, heterogeneity of vesicle populations, limited control over cargo loading, and the absence of standardized potency assays [116].

In addition, compared with synthetic nanocarriers such as silica–metal–organic framework nanoparticles, which demonstrate high (> 90%) loading efficiency and robust in vivo genome editing capability, EV-based systems still face challenges in achieving comparable scalability and reproducibility, underscoring the need for further optimization [117].

Despite these promising advances, important limitations must be acknowledged. Variability in engineering techniques, heterogeneity of vesicle populations, scalability of production, and the absence of standardized potency assays continue to impede clinical translation [118]. Concerns regarding the immunogenicity of engineered ligands, cost-effectiveness, and the regulatory frameworks required for clinical approval remain significant barriers [119]. Future research directions should therefore focus on the development of GMP-compliant production pipelines, the implementation of in vivo biodistribution tracking methods, and the integration of AI-assisted design platforms to accelerate the safe and effective translation of engineered sEVs into clinical hematology [120].

Bridge to clinical translation (human/clinical signals)

Human data supporting the therapeutic application of EVs, particularly small EVs, in hematologic diseases remain limited, with most available evidence derived from early-phase trials or non-hematologic indications [121]. Within hematologic settings, emerging clinical efforts are beginning to explore the feasibility of EV-based interventions. Notably, AML, an ongoing clinical study is evaluating non-engineered umbilical cord mesenchymal stem cell-derived exosomes (UCMSC-Exo) for the mitigation of chemotherapy-induced myelosuppression (NCT06245746), focusing primarily on safety, tolerability, and hematopoietic recovery rather than direct antitumor activity. Although this approach does not directly target malignant cells, it represents an important proof-of-concept for the safe application of EV-based therapies within hematologic patient populations.

In the hematopoietic transplantation context, mesenchymal stem cell-derived EVs have demonstrated immunomodulatory effects in graft-versus-host disease (GvHD), with preclinical and early translational studies reporting improvements in inflammatory regulation and survival outcomes; however, robust human data remain limited [122, 123].

Beyond hematology, the clinical development of EV-based therapies is more advanced and provides important translational insights. For example, BM-MSC-derived EVs (ExoFlo™) have been evaluated in a multicenter, randomized, double-blind Phase II trial in patients with COVID-19-associated ARDS, demonstrating a favorable safety profile with no treatment-related adverse events [124]. While efficacy outcomes in the overall cohort were modest, subgroup analyses suggested reductions in mortality and improvements in ventilation-free days, indicating context-dependent therapeutic benefit [124].

These early findings have supported progression to a large-scale Phase III randomized trial (EXTINGUISH ARDS; NCT05354141), reflecting increasing confidence in the clinical feasibility of EV-based therapeutics. Similarly, ongoing clinical studies investigating MSC-derived EVs for acute respiratory failure (e.g., NCT06002841) and nebulized EV delivery strategies (NCT05787288) highlight continued efforts to optimize administration routes and enhance tissue-specific targeting.

At a broader level, global analyses of EV-based clinical trials indicate that most studies remain in early developmental phases (predominantly Phase I/II), with substantial heterogeneity in EV sources, dosing strategies, and administration routes [121]. This lack of standardization, combined with incomplete mechanistic understanding and challenges in large-scale manufacturing, represents a major barrier to clinical translation and regulatory approval [121].

In contrast to non-engineered EVs, advanced bioengineered exosome platforms, such as anti-BCMA-targeted EVs for bortezomib delivery in multiple myeloma, remain confined to preclinical studies, with no available human data to support clinical implementation [13].

Collectively, current evidence suggests that EV-based therapies are generally safe and technically feasible in humans, particularly in early-phase clinical settings. However, their clinical translation in hematologic malignancies remains at an early stage, with limited disease-specific human data and a clear gap between preclinical innovation and clinical implementation. Addressing key challenges, including standardization of EV characterization, optimization of dosing strategies, scalability of production, and demonstration of consistent therapeutic efficacy, will be essential to enable the transition of EV-based therapeutics from experimental platforms to routine clinical use (Figure 1). To provide a structured overview of the current translational landscape, ongoing and recently registered clinical studies investigating EV-based applications in hematologic indications are summarized in Table 6.

Fig. 1.

Engineered extracellular vesicles (EVs) as multifunctional platforms in hematologic malignancies. This schematic provides an overview of major engineering strategies and therapeutic applications of extracellular vesicles (EVs), particularly small EV (sEV)-enriched populations, across key hematologic malignancies. The circular layout is organized by disease context (center: MM, AML, MPN, CLL) and corresponding EV-based engineering approaches (outer segments). Multiple myeloma (MM): Engineered EVs are applied for tumor antigen presentation (e.g., HSP70-anchored vesicles activating dendritic and T cells), siRNA delivery targeting oncogenic pathways, and antibody-modified EVs (e.g., anti-BCMA) for targeted drug delivery such as bortezomib, with additional effects on the bone marrow microenvironment. Acute myeloid leukemia (AML): Strategies include mesenchymal stem cell–derived EVs functionalized with c-Mpl ligands for targeted chemotherapy, PRIME EVs integrating immune checkpoint modulation with T-cell activation, RNA- and CRISPR-loaded RBC-derived EVs for genetic targeting (e.g., FLT3-ITD, miR-125b), and chemo-immunological cascade systems co-delivering chemotherapeutics and siRNA to induce immunogenic cell death. Myeloproliferative neoplasms (MPN): Engineered MSC-derived EVs incorporating targeting peptides (e.g., CXCR4, P-selectin) and loaded with USP5 inhibitors demonstrate enhanced homing to JAK2V617F-mutated stromal niches and suppression of aberrant proliferation. Chronic lymphocytic leukemia (CLL) and B-cell malignancies: EV-based approaches include antigen-presenting vesicles (e.g., CD154/EBV gp350), CAR-engineered EVs targeting CD19 for drug delivery, and co-loaded systems (e.g., rituximab/siPDK4) that restore CD20 expression and overcome therapeutic resistance. Across these disease contexts, engineered EVs function as both targeted delivery systems and immune modulators. However, key translational challenges, including scalability, reproducibility, cargo loading efficiency, and safety, remain to be addressed for clinical implementation

Table 6.

Ongoing and Recent EV-Based Clinical Studies in Hematologic Indications

Indication	EV application	EV source/type	Trial ID	Primary aim	Translational note
Myeloproliferative neoplasms (ET, PV, MF)	EV-based biomarker (senescence profiling)	Circulating EVs (patient-derived)	NCT06798805	Identify EV-based senescence biomarkers and prognostic signatures	Focuses on liquid biopsy/biomarker discovery, not therapeutic use
Hairy cell leukemia (HCL)	EV-mediated microenvironment analysis	Plasma-derived EVs	NCT06774677	Characterize EV-driven tumor–microenvironment interactions	Early-stage translational relevance; microenvironment-focused
Lymphomas, CLL, MM	EV-based diagnostic/prognostic profiling	Circulating EVs (including large EVs)	NCT06782854	Evaluate EVs as biomarkers of disease aggressiveness and response	Supports EVs in liquid biopsy frameworks
Acute myeloid leukemia (chemotherapy-induced myelosuppression)	Supportive EV therapy	UCMSC-derived EVs (non-engineered)	NCT06245746	Assess safety, tolerability, and hematopoietic recovery	Hematology-specific but non-antitumoral; supportive application
Relapsed/refractory B-cell acute leukemia	Engineered EV therapy (BiTE-EV)	MSC-derived engineered EVs (CD3/CD19 BiTE)	NCT06890494	Assess safety and preliminary efficacy	One of the few true therapeutic EV trials in hematology, early-stage

Theranostic EVs: real-time tracking and therapy

Theranostic EVs represent an emerging class of multifunctional platforms that integrate diagnostic and therapeutic capabilities within a single system. EV-based systems hold promise by enabling targeted delivery while simultaneously providing feedback on treatment response in real time [125]. Despite these promising attributes, the majority of EV-based theranostic and liquid biopsy applications remain in early translational stages and have not yet been fully integrated into routine clinical workflows.

The theranostic potential of EVs is increasingly evident in oncology, including hematologic malignancies. Metastasis remains a major driver of cancer mortality, and EVs released by tumor cells actively shape the tumor microenvironment, thereby influencing disease progression and treatment response [126]. Analyses of tumor-derived EV cargo, proteins, RNAs, and metabolites, can yield rapid and precise diagnostic signatures, while strategies that suppress or modulate the release of pathogenic EVs offer novel therapeutic avenues [127, 128]. Preclinical studies suggest that targeting tumor-associated EV release may limit clonal expansion before chemotherapy initiation, positioning EVs as dual-purpose agents for both disease monitoring and therapeutic modulation [129]. In this context, platforms such as imaging flow cytometry and microfluidic devices have been proposed to quantify EVs and circulating tumor cells (CTCs) at the single-vesicle level, while rapid diagnostic kits or self-testing strips are being developed to integrate EV analysis into routine practice [130].

Building upon these biological insights, nanotechnology-driven platforms have expanded the technological toolkit for EV-based theranostics. Dual imaging modalities such as magnetic resonance imaging (MRI) and positron emission tomography (PET) enable simultaneous high-resolution spatial mapping and sensitive biodistribution analysis [131, 132]. For instance, a ferritin–lactadherin fusion protein has been employed to label EVs for MRI, allowing longitudinal in vivo tracking; however, this approach also revealed potential biosafety concerns, including suppression of stem cell proliferation, that highlight the importance of rigorous safety assessments [133]. Complementary nuclear imaging approaches such as SPECT/CT using 111In–DTPA–labeled EVs have provided quantitative biodistribution data, confirming predominant accumulation in liver and spleen and underscoring the necessity of active targeting for hematologic applications [134].

In parallel, Raman-based methods, particularly surface-enhanced Raman scattering (SERS), have emerged as label-free tools capable of differentiating malignant from normal EVs with high specificity. Integration of SERS with machine learning algorithms has further improved diagnostic accuracy, enabling robust molecular fingerprinting beyond traditional biochemical assays [135–137]. Importantly, these optical platforms can be combined with other imaging strategies to minimize false positives and strengthen reproducibility, making them well-suited for precision diagnostics in hematology. A recent clinical study extended this approach to human plasma samples from patients with multiple myeloma (MM) and its precursor states, MGUS and aMM. sEV isolated from patient sera were characterized by size (~ 98 nm) and tetraspanin expression (CD9/CD63) and then profiled by Raman and SERS. Principal component analysis (PCA) successfully discriminated between MGUS, aMM, and symptomatic MM, with lipid-associated spectral signals serving as the main discriminants. This work provided the first clinical evidence that SERS-based profiling of EVs could stratify plasma cell dyscrasias, underscoring its potential as a minimally invasive diagnostic and monitoring tool [138].

Beyond diagnostics, targeted therapeutic applications of EVs are gaining momentum. In diffuse large B-cell lymphoma (DLBCL), iRGD-modified EVs carrying BCL6 siRNA selectively targeted αvβ3-integrin–positive lymphoma cells and markedly inhibited tumor growth in vivo, providing proof-of-concept for EV-mediated siRNA delivery in hematologic malignancies [139]. Likewise, PET-based labeling strategies using NOTA–64Cu or 68Ga–conjugated EVs have enabled quantitative tracking of exosome biodistribution in lymphatic and hematogenous routes, demonstrating accumulation in lymph nodes, lungs, and liver. These results highlight a clinically translatable theranostic scaffold directly applicable to blood cancers [140].

Exosome-based theranostic platforms provide a minimally invasive method to monitor disease progression and therapy responses in hematologic malignancies. EVs may also be promising theranostic targets for creating new immune-surveillance methods, depending on where they come from in the body [141]. Genetic engineering methods allow for real-time monitoring and tailored administration. Raman-based profiling, on the other hand, increases diagnostic accuracy by differentiating between the molecular fingerprints of malignant and normal cells. These approaches could work well together to improve early identification, tailored treatment, and long-term monitoring of blood cancers.

AI-Driven engineering of extracellular vesicles

Among current applications, the most mature and clinically relevant use of artificial intelligence (AI) in EV research lies in biomarker discovery, particularly through the integration of machine learning with EV-associated RNA profiling. Several recent studies demonstrate that AI-driven models can achieve high diagnostic accuracy when applied to multi-cancer datasets derived from circulating EVs.

Validated AI applications in EV biomarker discovery

For example, in a large multi-phase study integrating exosomal RNA sequencing and machine learning, Random Forest–based models trained on a refined 33-gene EV-derived signature achieved an area under the curve (AUC) of 0.887 in independent validation cohorts, with individual cancer-specific models reaching AUC values as high as 0.99 (lung cancer) and 1.00 (ovarian cancer) [142]. Notably, this performance was achieved across a heterogeneous cohort of more than 1,300 participants spanning eight cancer types, highlighting the robustness and scalability of EV-based AI diagnostics in clinically relevant settings [142]. Importantly, repeated resampling and cross-validation strategies confirmed the stability of these models, supporting their potential for real-world application.

Similarly, machine learning frameworks such as Random Forest, LASSO-integrated pipelines, and support vector machine (SVM)–based platforms (e.g., CancerSig) have demonstrated strong predictive performance in identifying stage-specific miRNA signatures, further reinforcing the utility of AI in EV-based liquid biopsy approaches. In hematologic malignancies such as diffuse large B-cell lymphoma (DLBCL), AI-enabled analysis of EV-derived miRNAs has also enabled the identification of clinically relevant regulatory networks, suggesting potential applications not only in diagnosis but also in patient stratification and therapeutic targeting [143–149].

Beyond transcriptomic profiling, AI-assisted analysis of biophysical EV properties has emerged as an additional diagnostic modality. Label-free approaches integrating electrokinetic measurements (e.g., zeta potential and conductance) with machine learning algorithms have demonstrated classification accuracies approaching 99–100% (AUC ≈ 1.0) using decision tree and gradient boosting models for EV and nanoparticle discrimination [150]. Even under reduced training conditions, advanced models such as quantum machine learning (QML) maintained predictive performance of ~ 92–94%, highlighting their robustness in data-limited settings [150]. These findings suggest that combining physical and molecular EV features with AI can further enhance diagnostic precision.

Collectively, these studies indicate that AI-driven EV biomarker discovery has progressed beyond proof-of-concept toward quantitatively validated diagnostic models with strong performance metrics. However, it is important to note that most current evidence remains derived from controlled cohorts and retrospective analyses. Variability in EV isolation methods, cohort heterogeneity, and lack of standardized MISEV-compliant datasets may still affect reproducibility and cross-study comparability. Therefore, while the diagnostic potential of AI-assisted EV profiling is compelling, further prospective validation in large, multi-center clinical studies will be essential to establish its routine clinical utility [142, 150, 151]. Importantly, no AI-enabled EV diagnostic or engineering platform has yet been approved for routine clinical use, underscoring the current gap between experimental performance and clinical translation.

AI-guided engineering of EV cargo and targeting

Beyond biomarker discovery, AI is increasingly being leveraged to address key engineering challenges in EV-based therapeutics. One major application lies in optimizing cargo loading efficiency. Machine learning models trained on experimental datasets, including electroporation parameters, lipid composition, and vesicle size, have been used to predict optimal conditions for nucleic acid encapsulation and controlled release. These approaches move beyond empirical trial-and-error strategies and enable rational design of EV cargo profiles tailored to specific disease contexts [152–155].

At the targeting level, AI-driven computational approaches are being used to design EV surface ligands, including peptides, antibodies, and aptamers. Techniques such as molecular docking, structural modeling, and generative algorithms allow in silico prediction of binding affinity and specificity prior to experimental validation. Widely used tools such as AutoDock (for molecular docking simulations) and deep learning–based structural prediction frameworks (e.g., AlphaFold-derived modeling pipelines) have been increasingly adapted to guide ligand–receptor interaction design and improve targeting precision in nanovesicle systems [156, 157].

Reinforcement learning frameworks further enable iterative optimization of ligand configurations, facilitating the identification of designs that maximize tumor selectivity while minimizing off-target interactions [158–161]. Collectively, these approaches highlight how AI contributes directly to EV engineering at both the cargo and surface-interface levels, thereby enhancing delivery precision and therapeutic efficacy.

Emerging AI applications and translational considerations

Despite these advances, several AI-relevant directions in EV engineering remain at a preclinical or proof-of-concept stage. For instance, engineered EV systems have already demonstrated that complex RNA and CRISPR-associated cargos can be selectively packaged and functionally delivered in vitro and in vivo, but these studies were achieved through molecular engineering rather than AI-guided design. In one representative study, CD9-HuR–functionalized exosome-like vesicles enhanced the loading of miR-155, miR-155 inhibitors, and engineered CRISPR/dCas9-related RNA cargos, leading to measurable target suppression in recipient cells and in mouse tissues, thereby establishing that EV-based RNA programming is feasible but still dependent on highly customized construct design and disease-specific optimization [162]. Similarly, the NanoMEDIC platform showed that extracellular nanovesicles can transiently deliver CRISPR-Cas9 ribonucleoprotein machinery, achieving efficient genome editing in hard-to-transfect cells and > 90% exon-skipping efficiency in Duchenne muscular dystrophy–derived skeletal muscle cells, with additional in vivo activity after intramuscular injection [163]. These findings are important because they demonstrate that EV-based gene-editing delivery is no longer purely conceptual; however, they also underscore that current success still relies on carefully engineered packaging systems, controlled experimental settings, and disease models that may not fully reflect the biological variability seen in clinical hematology.

From a translational perspective, this is precisely where AI may become most useful. Rather than being viewed as a validated therapeutic layer by itself, AI should currently be framed as a future optimization tool that could help prioritize cargo-loading parameters, predict vesicle stability, improve targeting-ligand selection, and model biodistribution before experimental testing. This is especially relevant because EV engineering still faces major bottlenecks, including low yield, batch-to-batch variability, cargo heterogeneity, and inconsistent isolation workflows. As also emphasized in recent methodological discussions, the clinical scalability of EV-based platforms remains limited by the lack of standardized datasets, variable characterization pipelines, and imperfect alignment between experimental studies and MISEV-compliant reporting frameworks [164]. Accordingly, the most realistic near-term view is that AI will likely contribute first to design refinement and decision support, rather than to fully autonomous EV therapeutic development. Overall, emerging studies support the biological feasibility of EV-mediated RNA and CRISPR delivery, but the clinical translation of AI-guided EV engineering will depend on better standardization, higher-quality training datasets, and prospective validation in disease-relevant models before these approaches can be considered truly mature for therapeutic application.

Comparison with synthetic delivery systems

A meaningful evaluation of engineered sEVs as delivery platforms requires comparison with established synthetic systems, particularly lipid nanoparticles (LNPs), which currently represent the clinical benchmark for nucleic acid delivery.

LNPs are characterized by high encapsulation efficiency, scalable and reproducible manufacturing processes, and well-defined physicochemical properties, which have enabled their successful implementation in mRNA-based therapeutics and vaccines [165]. In contrast, sEVs provide unique biological advantages, including superior biocompatibility, lower immunogenicity, and the ability to exploit endogenous intercellular communication pathways for more context-specific delivery [56]. These properties are particularly relevant in hematologic malignancies, where interactions within the bone marrow microenvironment play a critical role in disease progression and therapeutic response [166].

Nevertheless, compared to LNPs, sEV-based systems face important limitations, including lower and more variable cargo loading efficiency, intrinsic heterogeneity of vesicle populations, and significant challenges in achieving standardized large-scale production [167]. Furthermore, although LNPs rely primarily on passive distribution or engineered targeting ligands, sEVs utilize endogenous trafficking mechanisms, which may enhance targeting specificity but can also introduce variability and off-target biodistribution [165]. Importantly, these two delivery platforms should be considered complementary rather than competitive, as each offers distinct advantages depending on the therapeutic context.

While LNPs currently demonstrate greater translational maturity due to their scalability and regulatory readiness, sEVs may offer advantages in cell-type–specific delivery, immune modulation, and reduced toxicity [168, 169]. Future advances are therefore likely to involve comparative optimization and the development of hybrid delivery strategies that integrate the strengths of both synthetic and biologically derived systems.

Challenges and future directions

Analytical standardization and functional potency

Despite rapid progress, the clinical translation of extracellular vesicle (EV)–based therapeutics in hematology remains constrained by insufficient standardization at the analytical and functional levels, often reflecting incomplete adherence to MISEV2023 recommendations [170].

A critical limitation lies in EV characterization. Many studies rely primarily on particle size distribution or a restricted panel of surface markers, frequently omitting appropriate negative controls (e.g., ApoA1, GM130), thereby increasing the risk of co-isolation of non-vesicular contaminants. To address this, current guidelines recommend a multi-modal characterization framework incorporating transmembrane markers (e.g., CD9, CD63, CD81), cytosolic proteins (e.g., TSG101, ALIX), negative markers, and orthogonal validation methods including particle quantification and morphological imaging [8].

In parallel, inaccuracies in cargo loading quantification remain a major analytical concern. In particular, electroporation-based approaches may introduce aggregation artifacts that confound the distinction between true encapsulation and non-specific association, leading to overestimation of loading efficiency. This highlights the need for standardized, particle-normalized quantification strategies and orthogonal validation techniques.

Equally important is the evolution of potency assessment. Functional evaluation of EVs should move beyond descriptive readouts toward mechanism-based assays aligned with the intended therapeutic function. For example, CAR-engineered EVs should demonstrate antigen-specific cytotoxicity, while RNA-loaded EVs should validate target gene modulation at both transcript and protein levels. Emerging approaches, including pathway-specific assays such as NF-κB inhibition or apoptosis-rescue models, provide promising frameworks for defining potency in a reproducible and clinically meaningful manner [171, 172]. Collectively, the implementation of standardized analytical pipelines integrating characterization, cargo quantification, and mechanism-linked potency assays represents a prerequisite for reproducibility and cross-study comparability in hematologic EV research.

Manufacturing, quality control, and safety considerations

The transition of EV-based therapeutics toward clinical application is further limited by challenges in large-scale manufacturing and regulatory compliance. Production of clinical-grade EVs requires robust Good Manufacturing Practice (GMP)-compatible workflows, including standardized source cell selection, closed-system processing, and reproducible purification strategies. Variability in source cells, such as mesenchymal stromal cells (MSCs), CAR-T cells, or induced pluripotent stem cells (iPSCs), necessitates stringent control through master cell banks and validated culture conditions.

Downstream processing remains a critical bottleneck, with purification methods such as ultracentrifugation, size-exclusion chromatography, and tangential flow filtration requiring harmonization to ensure consistency and scalability. In addition, product release criteria must incorporate identity, purity, potency, sterility, and endotoxin testing, aligned with certificate-of-analysis (CoA) standards [173–175].

Safety evaluation extends beyond conventional sterility assessments. Comprehensive profiling of EV cargo is essential to exclude oncogenic or immunogenic components, residual process-related contaminants, and unintended bioactive molecules. This is particularly relevant in immune-engineered systems, where EV-associated signaling components may contribute to systemic toxicities. For instance, correlations between CAR-positive EV kinetics and early neurotoxicity (ICANS) in CD19 CAR-T recipients underscore the need for careful pharmacovigilance [101]. Furthermore, biodistribution and pharmacokinetics represent critical determinants of therapeutic efficacy and safety. Advanced imaging modalities, including PET and MRI using labeled EVs (e.g., ^64Cu-based tracers), enable quantitative tracking of tissue distribution and organ accumulation. Such approaches are essential for defining absorption, distribution, metabolism, and excretion (ADME) profiles and for mitigating off-target accumulation in organs such as the liver and spleen [176–178].

Clinical translation and ai-assisted optimization

Human clinical indicators are starting to come to light for instance, CD19 CAR⁺EVs have been identified up to two years post-infusion, and their concentrations may function as early indicators of ICANS. To facilitate clinical translation, prospective trials must include EV-specific objectives, such as validated liquid-biopsy panels (EV-RNA/protein signatures), standardized sample time points, and composite outcomes that combine treatment response and toxicity [100, 101]. At the same time, AI provides tremendous tools for accelerating development. AI can optimize ligand design and cargo composition in advance, forecast biodistribution patterns, and customize dosing techniques to patient-specific multi-omics EV profiles [15]. In the near future, practical applications may include AI-guided electroporation protocols for RNA/CRISPR loading, in-silico screening of targeting ligands like peptides or aptamers for leukemic clones, and adaptive phase 1/2 trial designs that escalate doses based on real-time EV biomarker data rather than delayed clinical readouts [16, 179–181].

Engineering constraints and translational optimization strategies

Despite technological advances, several engineering-related challenges continue to limit the clinical scalability and reliability of EV-based therapeutics. A major source of variability arises from the EV-producing cells themselves. Donor-dependent heterogeneity in primary cells, particularly MSCs, can significantly affect EV composition, immunomodulatory properties, and therapeutic potency. The adoption of standardized platforms, including clonal cell lines and iPSC-derived systems, offers a promising solution by enabling controlled, reproducible EV production [182].

Targeting specificity remains another critical challenge, especially in hematologic malignancies where malignant and normal cells share overlapping surface antigens. While ligand engineering strategies, including antibodies, peptides, aptamers, and CAR-based modifications, have improved targeting precision, off-target uptake in the liver, spleen, and reticuloendothelial system persists [183]. Emerging approaches such as dual-targeting strategies, microenvironment-responsive EVs, and computationally optimized ligand design may enhance selectivity, although further validation in clinically relevant models is required [184].

Efficient and scalable cargo loading also remains unresolved. While conventional methods such as electroporation are widely used, they often suffer from low efficiency and structural perturbation of EVs. Alternative strategies, including endogenous loading via genetic engineering, light-inducible systems (e.g., EXPLOR), and hybrid EV–nanoparticle platforms, offer improved control over cargo incorporation but require further standardization and scalability assessment.

From a translational perspective, these challenges should be viewed not as fundamental barriers but as engineering and process optimization problems. Evidence from early-phase clinical studies, particularly those involving MSC-derived EVs, supports the feasibility and safety of EV-based interventions. However, successful clinical implementation will depend on the integration of standardized production platforms, mechanism-based potency assays, and predictive models of biodistribution and safety into a cohesive translational framework.

Conclusion

EVs, particularly sEV-enriched populations, have evolved from passive byproducts to active regulators of hematologic malignancies, mediating tumor progression, immune evasion, and chemoresistance through complex intercellular communication networks. At the same time, advances in EV engineering have enabled their transformation into programmable delivery systems capable of transporting therapeutic RNAs, proteins, and gene-editing tools. Preclinical studies across multiple hematologic models demonstrate that these engineered platforms can achieve targeted delivery, immune activation, and modulation of disease-driving pathways. However, despite these promising developments, most evidence remains confined to experimental settings, with limited standardization of loading efficiency, biodistribution, and functional potency, underscoring a persistent gap between biological innovation and clinical translation.

From a translational perspective, EV-based technologies are progressing, but remain at an early clinical stage, particularly in hematologic indications where human data are still limited and often restricted to supportive or biomarker-driven applications. Critical challenges, including vesicle heterogeneity, donor variability, suboptimal cargo loading, targeting specificity, and large-scale manufacturing, continue to constrain clinical implementation. Importantly, these limitations appear to be largely technical rather than fundamental, and ongoing advances in MISEV-aligned standardization, GMP-compliant production, and AI-assisted design are expected to improve reproducibility and therapeutic precision. Collectively, EV-based systems represent a promising but still maturing platform, with the potential to contribute to next-generation precision oncology in hematology as methodological and clinical barriers are progressively addressed.

Author contributions

Said Murodkhon Murtazaev was involved in the conception of the study, data analysis, manuscript preparation, and monitoring. Moreover, Chou-Yi Hsu, Sally Hassan Zubair, Mohammad Abohassan, Pareshkumar N. Patel, Gunjan Singh, Vimal Arora, Priya Priyadarshini Nayak, Muhammad Shahid Iqbal, and Zahraa Khudhair Abbas contributed to the search for relevant manuscripts and the preparation of the manuscript. All authors contributed to the development of the search strategy, article selection, and manuscript preparation. Finally, all the authors reviewed and approved the manuscript.

Funding

This research did not receive any financial support from public, commercial, or nonprofit organizations.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.