To address numerous challenges currently faced by CAR-immune cell therapy in clinical applications, CAR-Exosomes have garnered increasing attention in recent years. Exosomes are a class of extracellular vesicles, 30 to 150 nanometers in diameter, possessing unique biological characteristics (52). They originate from intraluminal vesicles (ILVs) within multivesicular bodies (MVBs). These ILVs are subsequently released as exosomes when MVBs fuse with the cell membrane via exocytosis (13). The lipid bilayer membrane of exosomes, composed of sphingomyelin, cholesterol, and lipid raft-associated proteins, confers high stability and biocompatibility (53). Exosomes carry a rich array of bioactive molecules, including proteins such as transporters and signal transduction molecules, nucleic acid such as mRNA (54) and miRNA (55), making them important intercellular communication vehicles. they are involved in regulating various physiological and pathological processes, such as immune response (56), tissue repair (57), and tumor progression (58). Their endogenous structure, naturally carried molecular components, and low immunogenicity not only provide potential therapeutic effects but also position exosomes as promising diagnostic markers and drug delivery vehicles due to their excellent biocompatibility and stability (59, 60).

Building upon the established foundation of CAR-engineered immunotherapies such as CAR-T, CAR-NK, and CAR-M cells, researchers have pioneered CAR-Exosomes, an emerging therapeutic platform (61). Functioning as a “miniature” version of their parental immune cells, CAR-Exosomes inherit the biological characteristics and material transport capabilities of their cell of origin. Critically, the functional CAR structures expressed on their surface provide the ability to precisely recognize tumor-associated antigens, thereby enabling targeted intercellular communication and drug delivery (14, 18, 62). By leveraging the natural properties of exosomes, CAR-Exosomes exhibit excellent stability, tissue migration, and penetration capabilities, allowing them to reach tumor sites more effectively (63). Furthermore, as immune cell-derived vectors, CAR-Exosomes can also modulate the immune microenvironment and enhance immune responses in the anti-tumor process (18). Compared to traditional cell therapies, exosome-based therapies generally exhibit lower immunogenicity and a reduced risk of side effects (14), thereby improving safety. Furthermore, studies indicate that CAR-Exosomes can serve as a multifunctional platform for efficient drug loading, addressing diverse therapeutic needs and offering more adaptable and precise strategies for tumor treatment (14, 15). A comparative analysis of CAR-immune cells and CAR-Exosomes is presented in Table 2 .

CAR-Exosomes are derived from CAR-engineered immune cells through a standardized production process. Initially, T cells, NK cells, and THP-1 cells are genetically modified via lentiviral transduction to express chimeric antigen receptors, generating CAR-T, CAR-NK, and CAR-THP1 cells respectively. Following expansion, exosomes are isolated by ultracentrifugation (64), ultrafiltration (17), density gradient centrifugation (65), size exclusion chromatography (66), immunoaffinity capture (67), polymer precipitation (68), microfluidic techniques (69), and sorting (70), each have their respective advantages and disadvantages ( Figure 2 ). We have summarized these in detail, as shown in Table 3 . Of note is the significant potential demonstrated by emerging methods based on cell sorting technology (e.g., flow cytometry) (70, 71). Their outstanding characteristics include ultra-high specificity, high purity, low sample volume requirements, multi-parameter analysis capabilities, effective avoidance of non-target contaminants, and precise quantification accuracy. These advantages make them especially well-suited for obtaining highly purified exosomes and conducting detailed phenotypic analyses. Specifically, such cell sorting techniques also provide an effective approach for characterizing exosomes derived from CAR-T cells.

The ‘anti-third-party activity’ of exosomes is complex and multifaceted, encompassing both beneficial anti-tumor effects and significant safety considerations. The beneficial anti-tumor effects are as follows. First, recruitment and activation of immune cells: for example, a current study has shown that CAR-M-Exosomes not only target tumor cells but also, being rich in the chemokine CXCL10, can effectively recruit T cells and M1-type macrophages. Specifically, CAR-M-Exosomes significantly enhance T lymphocyte activation and migration, promoting their differentiation into CD8⁺T cells (18). Furthermore, CAR-M-Exosomes increase the proportion of M1 macrophages (18), which are known to have anti-tumor properties (86) ( Figure 4C ). However, further research is needed to fully understand the impact of CAR-Exosomes on other immune cells. Second, potential effects on other stromal cells: we hypothesize that if other cells in the tumor microenvironment, such as cancer-associated fibroblast or endothelial cells, express the same antigen as the target tumor cells on their surface, CAR-Exosomes could similarly recognize and directly exert a killing effect on them. Third, microenvironment modulation via cargo: CAR-Exosomes themselves may carry various bioactive molecules (e.g., cytokines, non-coding RNAs). We speculate that by releasing these substances in the TME, they could directly modulate various neighboring cell types and reshape the immunosuppressive microenvironment. It is noteworthy that this potential ‘anti-third-party activity’ also raises important safety considerations: beyond acting on the tumor site, whether circulating CAR-Exosomes might affect normal tissue cells (e.g., by expressing low levels of the target antigen or via off-target effects) leading to non-specific damage. This is a crucial question that needs to be clarified in future research.

CAR-Exosomes can inherit certain characteristics from their parent cells, such as CAR-T, NK, and macrophage cells. Consequently, modifying the parental cells represents a key strategy for engineering CAR-Exosomes. This approach includes employing genetic editing technologies to generate dual-target CARs, which can further mitigate tumor antigen escape and enhance the activation and function of CAR-T cells (88). Existing studies have validated the efficacy of various dual-targeting CAR-T cells, including CD19/BCMA CAR-T (89), BCMA/GPRC5D CAR-T (90), and B7-H3/CSPG4 CAR-T cells (91), which have laid a critical foundation for developing multi-targeted CAR-Exosomes. In a study focused on lung cancer treatment, researchers developed dual-target CAR-T cells simultaneously targeting MSLN and PD-L1. Exosomes derived from these cells can precisely target MSLN-positive tumor cells. Furthermore, the anti-PD-L1-scFv within the exosomes blocks PD-L1 in the tumor microenvironment, reducing the exhaustion of infiltrating T cells (16). Furthermore, this involves optimizing the CAR structure to enhance T cell activation, proliferation, persistence, and cytotoxic capacity (51). CAR-T cells can also undergo metabolic reprogramming; for instance, overexpression of glucose transporter type 1 in CAR-T cells induces metabolic reprogramming and enhances efficacy (92). These optimized traits may also be reflected in the secreted CAR-Exosomes. Therefore, modification of parental cells enables the engineering of their derived exosomes. In summary, by modifying the parental cells, researchers can systematically tailor the biological properties of CAR-Exosomes, leading to more potent and multifunctional therapeutic applications.

The advancement of CAR-Exosomes technology is being driven by multi-strategy synergistic engineering approaches. The newly developed ExoCAR/T7@Micelle intelligent delivery system successfully achieves spatiotemporally precise regulation of exosomal drugs through the integration of three core technologies: CAR targeting, T7 peptide-mediated membrane fusion, and thermosensitive micelle encapsulation, charting a course for next-generation smart delivery platforms. On the one hand, it utilizes T7 peptides to mediate efficient BBB penetration and employs CAR for precise tumor cell targeting. On the other hand, it incorporates a light-responsive amphiphilic copolymer, mPEG-TK-Ce6, to form a micellar core encapsulating the ferroptosis inducer RSL3, enabling controlled drug release upon specific light irradiation via Ce6-triggered TK bond cleavage. The experimental results demonstrate that this intelligent system exerts an anti-tumor effect by inducing ferroptosis in HER2-positive BCBM cells. It effectively addresses key limitations of conventional chemotherapy, such as poor BBB penetration, inadequate targeting, and severe side effects, offering a highly promising new strategy for precision cancer therapy with substantial clinical translation potential (17).

One of the primary challenges facing exosomes-based therapies is the low yield and efficiency of production. Exosome yields are typically less than 1 μg per milliliter of culture medium, whereas many studies require doses ranging from 10 to 500 μg of protein per mouse (100, 101). A clinical trial has indicated that each patient may require between 0.5 to 1.4 x 10^11 exosomes (102). To meet the demands of animal experiments and clinical trials for large quantities of exosomes, it is crucial to enhance the efficiency of their large-scale production. In recent years, researchers have focused on hollow fiber bioreactors and stirred-tank bioreactors to improve exosome yields (93, 103). For example, GMP-grade bioreactors, such as those from Quantum, have successfully produced large quantities of exosomes carrying Kras-G12D siRNA, demonstrating their potential to inhibit tumors in a pancreatic cancer model (104). Furthermore, another study reported a ~6-fold increase in exosomes concentration and ~3-fold increase in exosomes production rate when using MSC cells cultured on microcarriers in a vertical wheel bioreactor (similar to a stirred-tank bioreactor) with xenobiotic-free medium (102). Concurrently, techniques like differential centrifugation (105) and density gradient centrifugation (106) are commonly used for exosomes purification to increase purity; however, these methods have limited throughput and are insufficient for meeting the demands of large-scale clinical production. To address these issues, tangential flow filtration (TFF) presents itself as a promising purification technique (107). By allowing the culture supernatant to flow along the membrane surface, TFF reduces membrane fouling, and improves throughput and recovery, thereby enabling efficient separation, which is better suited for the large-scale production of exosomes.

The advancement of Artificial Intelligence (AI) is poised to accelerate the large-scale production of exosomes. In the future, machine learning models can be employed to analyze the transcriptomic, proteomic, and other omic data of high exosome-producing cell lines, predict their secretion capacity, and thus assist in cell line selection. Reinforcement learning algorithms can be utilized to dynamically adjust bioreactor parameters, including pH, dissolved oxygen, and stirring speed, enabling closed-loop optimization. Furthermore, AI can be applied for automated analysis of exosomes size and morphology, facilitating standardized characterization. Machine learning can also be employed to evaluate key indicators of different exosomes batches for quality control purposes. By combining AI for optimization across the entire production workflow, from cell preparation and media formulation to bioreactor operation and purification, it is possible to significantly reduce costs and improve production efficiency, thereby enhancing the accessibility of exosomes therapies. Therefore, integrating advanced bioreactor technology and tangential flow filtration with AI techniques holds promise for significantly increasing exosomes yield and quality, accelerating the progress towards clinical application.

Storage stability and activity retention are critical factors impacting the performance of exosomes-based therapeutics. Therefore, investigating effective preservation techniques to protect their biological activity, facilitate transportation, and enable clinical application is of paramount importance. Currently, common preservation methods include cryopreservation, lyophilization (freeze-drying), and spray drying (108). Among these, -80°C cryopreservation is considered the optimal long-term storage solution, effectively maintaining the stability and activity of exosomes (109, 110). Research has shown that GMP-grade targeted exosomes retained significant anti-tumor activity after six months of storage at -80°C (104), while exosomes derived from milk exhibited no change in physical properties after four weeks of storage at -80°C, with the stability of loaded drugs also being maintained (111). Furthermore, incorporating cryoprotectants such as human albumin and trehalose can effectively reduce membrane damage caused by ice crystal formation, further enhancing stability (109).

In the future, researchers can leverage AI models to analyze vast storage experimental datasets and automatically optimize cryopreservation conditions, including temperature, cryoprotectant type and concentration, and buffer formulations, thereby establishing optimal storage protocols for CAR-Exosomes. Based on omics data and characterization information, machine learning models can also predict the impact of different storage conditions on exosomes stability and quality, allowing for early identification of potential risks. Additionally, customized storage and transportation strategies can be designed, tailoring them to the source, cargo, and intended application of CAR-Exosomes, thereby improving inter-center and inter-batch reproducibility and enhancing the efficiency of clinical translation.

Despite promising preclinical results demonstrating significant anti-tumor effects of CAR-Exosomes, the precise mechanisms underlying their anti-tumor activity remain largely unelucidated. This includes how they activate the host immune system, kill tumor cells, and remodel the TME using signaling molecules. Furthermore, there is a lack of in-depth understanding of their in vivo behavior, including their biodistribution and interactions with target cells. Concurrently, the core bioactive components driving their activity, such as specific proteins, RNA, and lipids, have not been fully characterized. To systematically decode these complexities and promote clinical translation for targeting solid tumors, one can utilize multi-omics approaches (e.g., transcriptomics, proteomics, lipidomics), high-throughput functional assays, and animal studies to comprehensively elucidate exosomes composition, targeting properties, and the specific pathways involved in activating host immunity, killing tumor cells, and remodeling the microenvironment. For example, a study identified CXCL10 as the most enriched chemokine through proteomic identification of CAR-M-Exosomes and subsequent enrichment analysis, revealing that CAR-Exosomes exhibit significant immunomodulatory functions (18), providing important clues for further research on immune activation mechanisms. In addition, in vivo imaging and tracking techniques (112) such as BLI, MRI, PET, and SPECT can be used to study their in vivo distribution, duration, targeting, and metabolic pathways in detail.

With the advancements in AI technology, it is becoming possible to integrate the above multi-source data using AI machine learning and deep learning models to establish dynamic, multi-layered mechanism-of-action models. For instance, AI can be employed to analyze CAR-Exosomes proteomic data, as shown in a study that identified and validated serum exosomes proteomic features based on machine learning. Furthermore, AI can analyze changes in transcriptomic, proteomic, and metabolomic data following co-culture of CAR-Exosomes with tumor cells/immune cells, thereby revealing the signaling pathway regulatory mechanisms underlying CAR-Exosomes treatment of cancer. This approach can also clarify which signaling pathways the CAR-Exosomes utilize to affect immune cell activation. By inputting exosomes design parameters and synthesizing the results from previous animal experiments, it is also possible to predict their distribution in tumor tissues and other organs, and to devise exosomes administration protocols.

Clinical translation of CAR-Exosomes therapies faces significant safety challenges, necessitating systematic safety validation. Currently, research primarily relies on murine models, lacking systematic assessment of toxicology, pharmacokinetics (PK), and pharmacodynamics (PD) in non-human primates (NHPs). Before clinical application, the safety issues of primary concern for CAR-Exosomes therapies include immunogenicity, long-term organ toxicity, and off-target effects. To address these, high-potential translational strategies include: assessing the effects of CAR-Exosomes on the immune system, including monitoring the distribution of lymphocyte subsets and cytokine profiles; conducting systematic toxicological, PK, and PD studies in NHPs to comprehensively evaluate the safety of the exosomes and their potential risks. While one study demonstrated that intravenous administration of 3.85×10¹² human mesenchymal stem cell-derived exosomes did not elicit overt toxic reactions (113), the safety of engineered CAR-Exosomes in patients with tumors warrants further in-depth investigation. Simultaneously, conducting dose-escalation studies to determine the safe dosage of CAR-Exosomes for tumor treatment is crucial. Furthermore, utilizing whole-body imaging techniques like PET-CT (114) to track the distribution of CAR-Exosomes in animal models will allow assessment of off-target accumulation and potential toxicities. Moreover, designing CAR activity “switch” systems with drug or signaling control can reduce the risk of non-target toxicity when necessary, thereby improving overall safety. For instance, one study showed the use of protease-regulated CARs as a “switch” for CAR-T cells (115), deactivating CAR activity upon detection of potential toxic responses, ensuring treatment safety. Additionally, CAR-T-Exosomes can be engineered in combination with biomaterials to achieve localized or targeted delivery, mitigating toxicity to other organs. For example, fusion of CAR-T-Exosomes with lung-targeting liposomes can achieve drug enrichment within the lungs, reducing systemic side effects (16).

Furthermore, AI technology is expected to empower multi-level safety defenses. At the molecular design stage, AI can utilize tools such as AlphaFold to predict scFv off-target epitopes and immunogenicity, optimizing the stability of CAR structures and enhancing their safety. For in vivo behavior monitoring, intelligent quantification of exosome distribution and tumor enrichment efficiency can be achieved through the fusion of multi-modal imaging (PET/MRI/optical) technologies. Concurrently, deep learning can be used to construct toxicity prediction models by integrating observed toxic reactions and corresponding laboratory abnormalities. For instance, in CAR-T research, a study successfully predicted the occurrence of ICANS in 204 patients with follicular lymphoma or diffuse large B-cell lymphoma treated with axicabtagene ciloleucel by constructing a machine learning-based logistic regression model (116). CAR-Exosomes therapies can learn from the experiences of CAR-T and utilize AI technology for optimization. Through these comprehensive strategies, their safety is expected to improve significantly, accelerating the clinical translation process.

CAR-Exosomes therapy for solid tumors faces three major challenges: a limited selection of targets, the need for improved delivery efficiency, and the limitations of single-agent therapy. To address these bottlenecks, recent research focuses on three key areas. First, developing more targets for solid tumors is crucial. Given the heterogeneity of tumor cells, the development of CAR-Exosomes targeting multiple targets should also be explored, mirroring the success of dual-target CAR-T cell therapies (e.g., CD19/CD22) in refractory lymphomas (88). The future should see the development of multi-target CAR-Exosomes products. Second, integrating precision delivery systems to further enhance anti-tumor efficacy is essential. The application of inhaled PTX-loaded exosomes in lung tumors has significantly boosted drug concentrations (15). CAR-T-exosomes loaded with paclitaxel and fused with lung-targeting liposomes achieve drug enrichment in the lungs, enabling effective treatment of lung tumors and simultaneously reducing systemic toxicity (16). Third, the combination of CAR-Exosomes with multiple therapies may improve their therapeutic effect on tumors. For example, exosomes derived from γδ-T cells combined with radiotherapy can overcome radio-resistance in nasopharyngeal carcinoma cells significantly enhancing their therapeutic effect on nasopharyngeal carcinoma in vitro and in vivo (117). Furthermore, considering the potential of traditional Chinese medicine nanoparticles in immunotherapy (118), future research could explore the combined application of Chinese medicine nanoparticles with CAR-Exosomes. Additionally, multifunctional composite selenium nanoparticles have been proven to stimulate and enhance immune cell activity, particularly in antitumor treatments (119). These nanoparticles may be combined with CAR-exosome therapy. Therefore, this avenue warrants further attention, and efforts could be made to investigate the combined use of multifunctional composite selenium nanoparticles and CAR-Exosomes to achieve enhanced anti-tumor efficacy.

The rapid development of AI technology is becoming the core driver for strategies. In target discovery and screening: leveraging machine learning and deep learning algorithms, integrating multi-omics big data such as tumor genomics, transcriptomics, and proteomics, to predict and screen for highly expressed, highly immunogenic tumor-specific antigens. For instance, AI deep learning models have successfully identified the myeloid lineage markers CD86 and CSF1R as potential therapeutic targets for acute myeloid leukemia (120). Moreover, AI-driven simulation techniques, such as molecular dynamics simulations, can be used to optimize CAR-exosomes delivery strategies. For example, dynamic simulations of modified exosomes, predicting modification effects and influencing factors, can be performed to improve the targeting and stability of CAR-Exosomes within the tumor microenvironment. In terms of predicting efficacy and personalizing treatment, AI also holds significant promise. For example, studies have utilized CAR-T treatment datasets to train machine learning models, aiming to identify potential biomarkers of clinical outcomes. These efforts have successfully identified circulating lymphocyte subsets associated with clinical responses (121–123). In the future, AI technology can be used to integrate patient clinical data, imaging data, and gene expression profiles to build predictive models, thereby predicting the efficacy of CAR-Exosomes and providing a scientific basis for the development of personalized treatment plans.

Current research generally indicates that naked, un-engineered natural exosomes exhibit a relatively short half-life in the bloodstream. Specifically, after intravenous injection, their half-life is typically only 2 to 7 minutes. This results in their very brief presence in the body, often becoming undetectable within minutes to a few hours (124–127). This swift clearance is primarily attributed to the rapid recognition and phagocytosis by the host immune system, particularly the robust mononuclear phagocyte system (MPS/RES) found in the liver and spleen. Recently, researchers have explored various strategies to extend the half-life of exosomes. CD47 modification, by engineering exosomes to overexpress or incorporate this ‘Don’t Eat Me’ signal, allows them to bind to phagocytic cell surface SIRPα receptors, thereby inhibiting macrophage uptake and significantly prolonging plasma half-life for extended therapeutic action (128). Similarly, the conjugation of polyethylene glycol (PEG) molecules to exosome surfaces, creates a hydrophilic barrier that effectively lowers immunogenicity and non-specific binding, reducing MPS/RES clearance and, as demonstrated by some studies, extending half-life to approximately 10 hours (129). Beyond PEGylation, Polyoxazolines modification has also emerged as a promising option for stabilizing plasma exosomes, with research showing it can extend exosomes half-life six-fold within 6 hours post-injection (130), highlighting its immense potential. In addition, genetically modifying source cells to secrete exosomes pre-loaded with desirable membrane proteins (e.g., enhanced anti-phagocytic signals) may confer intrinsic stability to derived exosomes, which needs to be further explored.

AI holds immense potential for extending exosome half-life in the future. AI can simultaneously consider multiple variables such as PEG molecular weight, density, and linking chemistry, CD47 expression levels, exosome surface charge, and more. Through complex algorithms, it can identify the optimal combination to achieve the longest half-life with minimal toxicity. AI can also analyze vast amounts of in vivo pharmacokinetic data, biodistribution data, and immune response data to identify and quantify the contribution of different clearance pathways (e.g., MPS clearance by the liver and spleen, renal filtration) to exosome elimination. This helps in discovering key molecular features that lead to rapid clearance, thereby providing clear engineering strategies to circumvent these clearance pathways in the future.

To facilitate the therapeutic application of exosomes, the rapid development of exosome banking is crucial. The key to addressing this challenge lies in establishing and promoting a standardized process encompassing separation, purification, storage, and quality control, with a particular emphasis on maintaining exosome biological activity and integrity. Future strategies should focus on the following: First, integrating and optimizing multi-step combined separation strategies. Recognizing the limitations of single-method approaches, a multi-step approach can overcome drawbacks and enhance target purity and activity. For example, ultrafiltration or differential centrifugation can be used for initial concentration, followed by sophisticated purification techniques such as flow cytometry. Second, establishing a universally applicable standardized process. This involves developing standard operating procedures for sample collection, processing, separation, purification, quantification, QC, and storage, defining the optimal applicability of different methods and establishing evaluation standards for exosome purity, integrity, activity, and function. Third, on-going development and validation of novel and efficient technologies. This includes developing more rapid, gentle, high-throughput, highly specific techniques that can maintain exosome integrity and function, including but not limited to novel strategies for flow cytometry-based sorting, as well as AI and machine learning-driven sorting systems. Fourth, implementing rigorous multi-modal quality control. This requires the comprehensive application of techniques such as nanoparticle tracking analysis, transmission electron microscopy for morphological evaluation, exosome marker protein analysis by Western blotting, and RNA and lipidomic analysis, to ensure batch-to-batch consistency and biological activity of exosomes entering the bank. Fifth, further optimizing and developing innovative exosome storage strategies. This requires, in addition to existing cryopreservation methods, actively exploring new cryoprotective agents, optimization of lyophilization techniques, and stable storage strategies involving biomaterials. Furthermore, it’s crucial to conduct in-depth research to systematically optimize storage conditions and freeze-thaw limitations for different types of exosomes, and validate safer and more efficient cryoprotective agents.

In the future, we believe that AI can play a crucial role in resolving the challenges of exosome banking. AI can optimize separation processes through machine learning, automated systems, and the development of novel sorting technologies. It can utilize computer vision and machine learning to enhance image analysis, predict exosome bioactivity, and analyze batch-to-batch variations. Furthermore, AI can accelerate cryoprotectant screening and optimize storage conditions. Therefore, AI will be a significant driver in accelerating exosome banking construction and fostering the clinical application of exosome therapies.