Cytokines promote the survival, proliferation, differentiation and activation of lymphocytes. Interleukin (IL)−2,−15,−12,−21 as well as−18 improve anti-tumor function of NK cells and boost their proliferation in vitro and in vivo (30, 31). Here, we focus on the first two. A more detailed review can be found in Fang et al. (32), Lin and Leonard (33).

IL-2 was initially discovered as a T-cell growth factor more than 30 years ago and now is one of the most popular cytokines used to boost cytotoxicity of NK cells (30). However, a high-dose IL-2 therapy can lead to severe adverse effects, including vascular leakage and organ injury caused by activation of the vascular endothelium, where the IL-2 high affinity receptor, IL-2Rαβγ, is expressed (34). Furthermore, Treg cells also express high-affinity IL-2Rαβγ receptors and are preferentially activated by IL-2, inhibiting NK cell proliferation and cytotoxicity (35). These resulted in the development of alternative forms of IL-2. A new mutant of IL-2 called “super-2” has been constructed with increased affinity for the IL-2/15Rβ subunit present on NK cells, but lower affinity for IL-2Rα subunit (36). Alternatively, variants of IL-2, such as F42K, show decreased affinity in vitro for IL-2Rα (37). Another strategy is to create a recombinant fusion protein comprised of a cowpox virus encoded NKG2D binding protein with a mutated form of IL-2 that preferentially and potently stimulates IL-2 on cells bearing NKG2D only, without broadly activating IL-2Rα-bearing cells or inducing side effects in animal models (38).

A superior alternative to IL-2 is IL-15, which preferentially stimulates memory CD8⁺ T cells, and immature and mature NK cells. IL-15 receptors consist of IL-15Rα, IL-2/15Rβ, and γc. IL15Rα expressed on antigen-presenting cells such as DCs and monocytes can present IL-15 in trans to IL-2Rβγc receptors expressed on NK and CD8⁺ T cells without activating Tregs (39). IL-15 is the principal γc family cytokine in the cytotoxicity, homeostasis and development of NK cells (30). In the first clinical trial where recombinant human (rh) IL-15 was infused to patients with metastatic malignancies, NK cell proliferation was observed in patients (40). Although two patients showed the clearance of pulmonary lesions, there were no objective responses. In addition, the clinical use of rhIL-15 might be compromised by its short half-life and poorly tolerated by patients (MTD: 0.3 μg/kg/day). Interestingly, rhIL-15 was found to modulate the homeostasis of NK cells, memory CD8⁺ T cells and γδ T cells, but not Treg cells (40). Moreover, a most recent phase I clinical trial reported that subcutaneous administration of rhIL-15 significantly elevated the tolerated dose of rhIL-15 in patients with refractory solid tumor (MTD: 3 μg/kg/day). Subcutaneous administration of rhIL-15 remarkably promoted the proliferation of circulating NK cells and CD8⁺ T cells, especially CD56^bright NK cells (41). The biological function of rhIL-15 allows it to be used in combination with other modalities. Indeed, there are several clinical trials (NCT03759184: obinutuzumab; NCT02689453: alemtuzumab; NCT03388632: nivolumab and ipilimumab; NCT03905135: avelumab) underway investigating rhIL-15 in combination with antibodies.

ALT-803 is a superagonist of IL-15 comprising an IL-15Ra fused to IgG1Fc which is bounded to IL-15 mutein (N72D). This special compound is designed to simulate the typical trans presentation of IL-15 and has a much longer half-life (25 h vs. < 40 min), which may lead to enhanced ADCC (42, 43). In the first phase I clinical study with ALT-803, 33 patients who relapsed after allogeneic hematopoietic cell transplantation (allo-HCT) received ALT-803 intravenously or subcutaneously. ALT-803 is generally well-tolerated, as no severe dose-limiting toxicities (DLT) and graft-vs.-host disease (GVHD) were observed in both cohorts. Nineteen percent of the subjects showed clinical benefit, with 1 complete remission lasted for 7 months. All patients in this study exhibited increased levels of circulating CD56^brightNK cells and CD8⁺ T cells, especially those administered subcutaneously (44). Considering its prolonged half-life and potent activating power for CD8⁺ T and NK cells, ALT-803 is being evaluated in a range of ongoing clinical studies in combination with antibodies (NCT02523469: nivolumab, NCT02384954: rituximab) or adoptive transfer of NK cells (NCT01898793, NCT02782546). In a recent Phase 1b clinical trial, ALT-803 combined with anti-PD-1 antibody nivolumab was well-tolerated in non-small cell lung cancer (NSCLC) patients, and 6/21 of patients showed an objective response. Notably, among 11 subjects previously treated with nivolumab alone and relapsed, 3 subjects exhibited partial response, suggesting the addition of ALT-803 may reinforce the anti-tumor activity of antibodies (45). In clinical studies combining ALT-803 with adoptive transfer of NK cells (NCT01898793, NCT02782546), subcutaneously injected ALT-803 is expected to enhance the in vivo activation, survival and expansion of donor derived NK cells without severe adverse events (46). Given the fact that ALT-803 has significant anti-tumor activity and did not aggravate side effects associated with nivolumab or Bacillus Calmette Guerin (BCG) (47), meaningful results are promising to be obtained in these underway trials.

Monoclonal antibodies (mAbs) possess strong inherent efficacy that suits cancer immunotherapy as they have been shown to directly and/or indirectly promote the roles of NK cell in vivo. In this section we will discuss tumor-specific antibodies, specific killer engagers and antibodies targeted inhibitory receptors (on NK cells).

Adoptive transfer of autologous NK cells expanded ex vivo for treatment patients with lymphoma, colon cancer, breast cancer and lung cancer have been tested in a range of clinical trials. Only very limited antitumor effect was observed (83–85). The major reason was that the inhibitory receptors on autologous NK cells matched self MHC class I presented on tumor cells, and this “self” recognition signals subsequently inhibited the activation of NK cells (86). Besides, autologous NK cells derived from cancer patients were actually in an immune suppression state with impaired functions, making these cells difficult to exhibit antitumor capability. The first piece of evidence showing that NK cells had a clinical benefit was reported in 2002 (87). It has comfirmed that donor-vs.-recipient NK cell alloreactivity, which was mainly resulted from KIR ligand incompatibility, could avoid relapse and graft rejection without GVHD in AML patients receiving HLA (Human leukocyte antigens) mismatch donor hematopoietic transplantation (87). Later, this strategy was used in adoptive cellular immunotherapy of ex vivo activated allogeneic KIR/KIR ligand mismatched NK cells derived from PBMC into AML patients (88). Subsequently, alloreactive PBMC derived NK cells have been widely investigated as an immunotherapy in clinical trials of hematologic malignancies, as well as in trials of solid tumors including melanoma, breast cancer, ovarian cancer, neuroblastoma, renal cell carcinoma, colorectal cancer, and hepatocellular cancer (89, 90). The rules for infusion of allogeneic NK cells as cancer treatment have been reviewed by Leung Wing (91). For manufacturing large numbers of alloreactive NK cells prior to reinfusion, cytokines such as IL-2 and IL-15 were initially used for ex vivo culture and expansion of primary NK cells from peripheral blood (PB) (92). However, it was difficult to obtain large numbers of functional NK cells by only exposure to cytokines. Subsequently, artificial antigen-presenting cells (aAPCs) were developed as feeder cells for expanding NK cells ex vivo. Our group developed K562-based aAPCs with CD137L-IL-21, and found that they facilitate the log growth phase of NK amplification without indication of senescence in vitro culture (Invention patent NO. ZL201110075736.1). Using the aAPCs with CD137L-IL21 as feeder cells, a closed, large-scale and automated bioreactor under GMP conditions can be successfully used for manufacturing large numbers of activated NK cells, which has significantly promoted clinical trials for the application of NK cell-based immunotherapy.

The second way to produce large numbers of cytotoxic NK cells for immunotherapy is through a clonal NK cell line, such as NK-92, KHYG-1, HANK-1, NKG, NK-YS, YTS, YT, NK3.3, and NKL, all of which can expand and propagate easily in culture medium (93). Among them, NK-92 is the only cell line approved by the US FDA for use in the clinical trials, though these EBV-transfected cells must be irradiated before adoptive transfer to prevent propagation in patients (94). NK-92 cells can easily proliferate in flasks, gas-permeable bags, or bioreactors in compliance with GMP, with a doubling time between 24 and 36 h to produce effective clinical-grade NK-92 effectors (93). Although the safety and tolerability of NK-92 cells have been tested in several Phase I clinical trials, limited data exist on their clinical benefits (94–96). In a recent dose escalation trial (NCT00900809), the feasibility of adoptive transfer of NK-92 cells was assessed in 7 patients with refractory/relapsed AML. None of the patients experienced DLT, while no one had complete remission. Besides, no significant changes were observed in either counts or activity of NK cell, CD8+ T cell, CD4+ T cell, T regulatory cell and myeloid-derived suppressor cell (MDSC) (97). It has been indicated that the limited clinical efficiency of NK-92 cells is mainly due to the lack of CD16 expression and their short longevity in vivo (32). To overcome these barriers, high affinity natural killer (haNK) cells, a variant of NK-92 cell line engineered to CD16A (158VV) receptor and endogenous IL-2, has been successfully developed. This variant can mediate ADCC as well as bypass the need for exogenous IL-2 in culture (98). The feasibility of haNK cells is currently under investigation in patients with solid tumors (NCT03027128). Moreover, target-activated NK-92 (taNK) cells, another engineered variant of NK-92 cells modified with CARs, have been designed to target TAAs-expressing tumor cells (99, 100). The stable CAR expression and activity of these cells have been demonstrated by HER2.taNK (HER2-specifc target-activated NK), a cell line that is now being tested in patients with recurrent HER2-positive Glioblastoma (NCT03383978).

The third source of NK cells for tumor lysis are NK cells differentiated from human pluripotent stem cells (hPSCs), including hematopoietic stem/progenitor cells (HSPC) from umbilical cord blood (UCB), induced pluripotent stem cells (iPSCs), and human embryonic stem cells (hESCs) (101–103). NK cells derived from hPSCs can be utilized as an allogenic product. The protocols for producing NK cells from hPSCs often involve separating or generating CD34+ hematopoietic precursors and culturing in the presence of NK cell maintaining cytokines (IL-3, IL-7, IL-15, SCF, FLT3L) to further stimulate differentiation and growth of NK cells (104). These NK cells can be expanded by coculturing with aAPCs, which resulted in a 2–3 log expansion that lasted for more than 2 months (105). NK cells differentiated from stem cells have a phenotype close to primary NK cells, a stronger proliferative ability and a homogenous, genetically defined population which meets the requirements of their clinical applications (103, 104, 106). Recently, it has been demonstrated that long-term cryopreservation (1–10 years) had no effect on the expansion potential and effector function of UCB-derived NK cells, which made UCB an emerging source of NK cells. Particularly, there are over 600,000 UCB -units stored globally (107). The safety and tolerability of some UCB-derived NK cells have been tested in phase I clinical trial against AML and myelodysplastic syndromes (108).

In order to obtain stronger tumor killing activity of transferred NK cells, immune-stimulatory molecules such as antibodies and cytokines are often used in combination with NK cells in clinical studies (89, 109). Moreover, considering that NK cells can modulate host immune system against cancers, and hence function as immune adjuvants, they are able to boost the efficacy of anticancer therapy when combined with traditional treatments such as chemotherapy. In several clinical trials on NK cell-based immunotherapy, patients received chemotherapy prior to NK cell infusion (www.clinicaltrials.gov). Particularly, several drugs used in cancer chemotherapy could stimulate expression of activating ligands on tumor cells, which makes tumor cells more sensitive to NK cell-mediated cytotoxicity. Some of these drugs may also promote the recruitment of NK cells to the tumor sites by increasing secretion of selected chemokines (110). Among them, lenalidomide seems to be the most active drug which overcomes suppression of human NK cell anti-tumor functions by TME-associated IL-6 and TGF-β and lowers the threshold of MICA for NKG2D-mediated NK cell activation and augments nanoscale rearrangements in cortical actin at the NK cell ADCC immune synapse (111, 112). In clinical settings, the clinical trials of expanded and activated NK cell adoptive transfer are always in conjunction with lenalidomide treatment (NCT02573896, NCT02280525, NCT02481934, NCT02525250).

Unlike immunostimulatory strategies which involve the delivery of molecules to assist NK cells, genetic modification strategies induce changes in the genetics of NK cells directly, leading to far-reaching and sustained changes to the cells (113). Among them, genetic modification of NK cells with CAR constructs has drawn increasing attention. CAR-NK cells can be produced from different sources of NK cells including primary NK cells, NK cell lines, and HPSCs. CAR-NK has adopted the basic structural framework of CAR-T, that is, chimeric antigen receptors mainly composed of extracellular, hinge, transmembrane and intracellular domains, as well as the transfection methods (100, 114, 115). The extracellular domain can bind tightly to tumor-associated antigens expressed on the surface of tumor cells, which determines the specificity of CAR structures. Single-chain variable fragments (ScFvs) are the most commonly used ectodomains for CARs. The hinge domain is the connecting sequence between the extracellular domain to the transmembrane domain, which endowed CAR with sufficient orientation and flexibility to bind to tumor antigens and are expected to impact the CAR-NK activities. The transmembrane domain lies between the hinge and the intracellular signaling domain, including CD3ζ, HLA-A2, or CD28 molecules. The structure of the intracellular signaling domain determines the intensity of the CAR-NK activation signal, which contains the immunoreceptor tyrosine-activated motifs (ITAMs). The majority of current CAR endodomains contain an activation region derived from CD3ζ, which is the most classical intracellular domain including three ITAMs. In order to increase the proliferation and cytotoxicity of CAR modified effector cells, co-stimulatory protein receptors such as CD28, 4-1BB, CD134, ICOS are added to the cytoplasmic tail.

The first CAR used in NK cells is a CD4-CD3ζ (CD4ζ) fusion receptor, which was reported by Tran et al. (116). In this study, the CD4ζ chimeric receptor is biochemically and functionally active and can guide human NK cells efficiently to kill either HIV-infected CD4⁺ T cells or NK-resistant tumor cells expressing gp120 in vitro, indicating CAR structure can be successfully expressed on NK cells, leading to efficient retargeting (116). Subsequently, more attempts were taken to enhance the anti-tumor ability of NK cells. Uherek et al. modified NK-92 cells with ErbB2-specific CAR, and found these cells show specific cytotoxicity in animal models (117). While, Imai et al. evaluated the efficacy of primary NK cells expressing CD19 CAR against autologous leukemic cells in vitro, and demonstrated these cells can bypass inhibitory signals (118). To date, CAR-NK cells have been evaluated for the treatment of hematological cancers and solid tumors in preclinical studies with numerous good results [reviewed in (32, 36, 100)]. On the basis of these findings, the clinical translation of CAR-NK cells has witnessed significant interests (see Table 1 for details of ongoing clinical trials). Although CAR-NK therapy is still under clinical evaluation, NK cells possess several advantages over T cells in being engineered to express CARs and used for cancer treatment. First, NK cells are easy to be isolated and have a relatively short lifespan. Therefore, the risk of overexpansion of transferred CAR-NK cells in patients is relatively low. Second, the cytokines secreted by NK cells mainly include IFN-γ and GM-CSF, which are relatively safer than those released by activated CAR-T cells, e.g., TNF-α and IL-6. In particular, the proinflammatory cytokines produced by CAR-T cells may cause life-threatening cytokine release syndrome (CRS), a most common and severe side effect of CAR-T therapy (119). Third, CAR-NK cells could trigger the lysis of target cells in both CAR-dependent and CAR-independent manners, which further potentiating their killing activity. In addition, CAR-NK therapy is expected to be less expensive, considering that NK cells can be derived from PBMCs, NK cell lines and hPSCs. However, T cells used for CAR-T therapy are required to be autologous (119, 120).

Despite of the significant advantages and great potentials, there are still challenges that should be addressed carefully to prevent hindering the clinical application of CAR-NK therapy. First, design of the optimal structure of CARs on NK cells has not been systematically studied yet. The position of the CAR-binding epitope and its distance to the CAR-NK cell surface are expected to affect the binding to the antigens, the optimal formation of immune synapses and CAR-NK cells activation (100). Empirical observations from CAR-T cells indicate that the structure of the hinge and transmembrane domains can affect CAR specificity and activity. In addition, there is a differential requirement for costimulatory signals between the established NK-92 cell line, primary NK cells and hPSC-derived NK cells. The study with three kind CD19-specific NK-92 cells, whose CARs harbor the same anti-CD19 scFv and hinge region, but linked to human CD3ζ, composite CD28-CD3ζ or 4-1BB-CD3ζ extracellular domains, respectively, showed that 4-1BB-CD3ζ-based CD19-specific NK-92 cells were less effective than CD3ζ or CD28-CD3ζ in cytotoxicity and cytokine production (121, 122). While for CD19-specific primary NK cells expanded ex vivo, inclusion of 4-1BB or 2B4 costimulatory domains together with CD3ζ was shown to promote both cytokine production and specific cytolysis (116, 123). It is also notable that, recently, Li et al. expressed CAR constructs which were specifically tailored to promote NK cell activation in iPSC-derived NK cells. They showed that such constructs containing the transmembrane NKG2D and intracellular 2B4 domains exhibit superior performance compared to CAR constructs designed for T cells (124). Given the unique activating receptor repertoire and signal-activating proteins of NK cells, customizing NK-based CAR structures will have great potential for improving CAR-NK function.

Second, although optimization of transduction methods in the last few decades has led to higher efficiencies, retroviral transduction is dependent on the existence of an active cell division, which weakens the efficiency of this strategy with non-activated NK cells. Genotoxicity of insertional mutagenesis is still a potential safety concern for lentiviral vectors. Furthermore, NK cell death associated with these procedures has limited the use of genetically engineered NK cells. Only few investigations have documented results regarding the capability of NK cells to function successfully after viral transduction. The innate properties that characterize NK cells may result in their reduced efficacy after viral transduction and apoptosis. As part of the immune systems innate defense mechanisms, pattern recognition receptors (PRRs) are used to identify pathogen associated molecular patterns (PAMPs). These procedures are likely involved in triggering a cascade of cellular events that lead to apoptosis of NK cells after viral transduction (125). Nevertheless, RNA engineering gave NK cells higher survival rates and transfection efficiency compared to viral gene transfer. One study found that NK cells differentiated from gene modified human hematopoietic stem cells have high gene transfer efficiency and viability of CAR-NK cells (126).

Third, albeit the primary targets of CARs are antigens expressed on abnormal cells, most currently available CARs are directed against non-mutated autoantigens that are differentially expressed on cancer cells. This may result in non-specific binding of CARs to antigens expressed on normal tissues. Therefore, antigen-positive healthy tissues are at risk of accidental injury induced by off-target activity. Anti-CD19 T cell therapy for B cell malignancies has shown strong off-target toxicity against normal CD19-positive B cells (127). There is no clinical data to assess the extent of off-target toxicity of CAR-NK, but identification of specific tumor targets is the basis for a safe and effective CAR construct. To increase the specificity and safety of CAR molecules, the experience from CAR-T studies can be useful. Effective reduction of cytotoxicity occurs if two individual structures, or two extracellular domains, are combined in a tandem form to produce a bi-specific CAR molecule (128–130). Importing a suicide gene, such as inducible caspase-9 (iCasp9) (131), facilitates control of dosage for CAR activity and provides another strategy to mitigate toxicity. A similar concept was described with a split CAR construct that requires dimerization of a small molecule to form a functional unit, further increasing safety (132).

Another safety concerning is that the infused expanded allogenic NK cells derived products may be contaminated by T or B cells, theoretically resulting in GVHD or post-transplant lymphoproliferative disorders, respectively. One study reported that donor-derived, IL-15/4–1BBL-activated NK cells infused early after major histocompatibility–matched, T-cell–depleted peripheral blood stem cell transplantation in patients could cause acute GVHD (133). They also demonstrated that GVHD was highly frequent in matched unrelated donor transplants and linked to higher CD3 chimerism. This later aspect refers to the establishment of an optimal clinical-grade protocol for purification of activated CD56⁺CD3⁻ NKs. This protocol could be safer for patients even in the presence of transfused third party NK cells modified with CARs. In any case, it is encouraging to note that clinical studies of haploidentical and UCB-derived NK cell infusions in hundreds of patients with both hematologic and solid malignancies have not reported a higher risk of GVHD (32).

EVs are used to depict three types of vesicles with lipid bilayer membrane. Exosomes are the smallest membrane vesicles (40–100 nm) that are secreted by most cell types by multi-vesicular bodies (MVBs) fusing with the cell membrane. Microvesicles (MV, 50 nm-1 μm) are vesicles that a little larger than exosomes, budding outwards from the plasma membrane. Apoptotic bodies (50 nm-5 μm) are the third kind of EVs, which are generated from cells undergoing apoptosis, by an outward blebbing way. In this review, we defined “EVs” as small extracellular vesicles with nanometer size. In transmission electron microscopy images, EVs usually appear as cup-shaped entities (Figure 2A). EVs express both common biomarkers, such as tumor susceptibility gene 101 protein (TSG101) and tetraspanins (CD9, CD63, CD81, and CD81), and specific proteins based on their cell origin (134). For instance, EVs derived from human NK cells (NK EVs) contain not only typical NK markers (e.g., CD56), but also lytic proteins (e.g., FasL, perforin).

EVs are involved in a number of key cellular processes due to their ability to transfer bioactive cargos (i.e., nucleic acids, lipids, and proteins) to recipient cells or by activating signaling pathways in target cells (135–138). Therefore, they are considered as potential alternatives to cell-based therapeutics, with characteristics possibly superior to cells in some aspects. First, since EVs are nano-sized, they can diffuse through tissues passively, partly due to the leaky vasculature of the tumor and subsequent enhanced retention effect (139). Second, although the acidic environment of solid tumors caused by hypoxia-ischemia increases immunosuppression (140, 141), it is beneficial to promote the fusion between EVs and tumor cells (142). Third, these “nano bullets” have an advantage in their ability to traverse biological barriers such as the blood-brain barrier (BBB) and the blood-tumor barrier (BTB), which has provided a completely different methodology compared to the cell-based approach (143–145). Moreover, EVs are stable at −80°C, lasting up to 12 months (146). This makes EVs easy to store and readily accessible as “off-the-shelf” supply.

The first study on NK EVs study was conducted by Lugini et al. (147). In this study, it was demonstrated that isolated NK EVs from purified NK cell culture exerted a cytotoxic effects on malignant hematologic cell lines, but not on solid tumor cells like breast carcinoma SKBR3 (147). Later, activated NK EVs, obtained from large scale isolation of in vitro expanded NK cells, showed significant cytotoxic effects on a series of cancer cell lines, including ALL, neuroblastomas, and breast carcinomas (146). These findings suggest that activated NK cells may produce EVs with higher immune activity compared with naïve NK cells. The antitumor effect of NK EVs is further supported by two recent in vivo studies. In one study, NK EVs were isolated by density gradient ultracentrifugation from NK-92 cells, and then injected into melanoma locally in a mice model. As a result, the progression of tumors treated with NK-92 EVs were significantly inhibited (148). In another study conducted by the same group, NK EVs prepared in the same way were used for the treatment of glioblastoma xenograft tumors. The intravenous administration of NK EVs not only significantly suppressed tumor growth, but also exhibited tumor-specific accumulation and an ability to cross the BBB. Encouragingly, this tumor-specific accumulation persisted for more than 5 days (149).

Fais et al. reported that NK EVs could trigger the death of target cells by two distinct mechanisms (137, 150), including ligand-receptor interactions (151, 152) and plasma membrane fusion (142) (Figure 2B). Furthermore, caspases, the initiator and executioner, were involved in the cell killing mediated by NK EVs (Figure 2B). Wen et al. extrapolated that the killing mechanisms of NK EVs can be classified into three groups, including perforin-granzyme mediated entry, receptor-ligand mediated interaction and granulysin mediated action (153). However, the underlying mechanisms of specific killing of tumor cells mediated by NK EVs remains unclear. NK cells recognize target cells for killing by a “self-missing” mechanism, but there is no evidence to prove that EVs can activate signaling pathways like living immune cells do. One possible reason why NK EVs can have a targeting effect is that the acidic microenvironment of solid tumors can promote the fusion of exosomes. In the study by Lugini et al. mentioned above, the secretion of NK EVs was stable and had a killing effect on activated PBMC cells in vitro, suggesting that NK EVs behave like a large “reservoir” to regulate the balance of immune cells (147). This mechanism is still not being clearly delineated. However, its potential for application in autoimmune diseases that result from excessive immune activation is obvious, as well as its potential use for elimination of side effects of immunotherapy such as cytokine storms is also worth exploring.

Notably, even though their anti-tumor activity has been demonstrated in vitro and in vivo, NK EVs have no impact on normal cells (147). In addition, due to the protective barrier offered by a natural vesicle structure, NK EVs might be used as a promising delivery system, which can be loaded with various therapeutic substances, e.g., genes and proteins. Particularly, the surface of EV membrane can be further modified by other functional groups.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.