Ultracentrifugation is the most popular and widely accepted gold standard for isolating exosomes [60]. It includes two types: differential ultracentrifugation and density gradient ultracentrifugation. Differential ultracentrifugation separates exosomes from other components by using different centrifugal forces and time cycles during the centrifugation process, mainly because different molecules have different particle sizes [63]. In density gradient ultracentrifugation, a continuous or discontinuous density gradient in a centrifuge tube is established with a specific medium, followed by the addition of a sample to the top of the medium to distribute different molecules in different fractions through gravity and centrifugal force fields, thereby separating exosomes and other components [63]. Ultracentrifugation requires fewer reagents and less operator expertise. Still, it may induce aggregates comprising various EVs and may be contaminated by lipoproteins with similar density [61, 62].

Ultrafiltration uses membranes with precise pore sizes to separate particles in a predetermined size range. Ultrafiltration alone or combined with ultracentrifugation and gel filtration chromatography is effective for exosome isolation [82]. Compared to traditional ultracentrifugation, ultrafiltration improves the purity and integrity of the separated vesicles while taking less time. The adsorption of exosomes on the filter membrane can result in decreased yield. Orifice holes may lead to deformation and rupture of large vesicles or platelets, which may skew the results [64, 65, 77].

SEC can be employed to isolate exosomes depending on the porous stationary phases that classify macromolecules and particles according to their size. Components with smaller hydrodynamic radii can enter these pores, resulting in extended retention durations, whereas exosomes and other substances with large hydrodynamic radii are excluded from the pores, thereby having a shorter retention time [63]. To ensure that the resolution in SEC meets the experimental requirements, the analysis is usually carried out at relatively low flow rates by using a rather long column (or multiple columns in series), making the separation process time-consuming. To address this shortcoming, several companies have developed commercial kits that enable higher separation efficiency and faster analysis [74].

F4 is a hydrodynamic-based technique to separate exosomes by diameter differences. Fractionation occurs in a rectangular flow field with migratory flow moving along the axis of the flow field, whereas sample retention depends on the rate of the secondary flow (cross flow in F4). The lateral flow acts as an external field to induce movement of the sample towards the flow field walls. Due to diffusion, the sample is distributed at different locations on the flow field walls, which can be separated according to the size of the sample particles. The distribution layer formed by the smaller particles has a higher average height above the stacking wall than the larger particles and is eluted earlier, enabling the isolation of exosomes [68]. Combined with an assay system, this method allows the rapid isolation and characterization of exosomes [63].

HFD is mainly used to isolate urinary extracellular vesicles by diffusion of solutes across a cellulose ester dialysis membrane with a defined molecular weight cut-off. The solvent will also pass through the dialysis membrane under hydrostatic pressure, thereby achieving filtration-concentration dialysis [71]. The advantage of HFD is that it can efficiently pretreat and concentrate samples and does not involve the operation of highly specialized equipment, making it a convenient method. HFD is significantly better for a wide range of sample volumes than differential centrifugation [70].

Polymer-based precipitation utilizes the interaction of polymers such as polyethylene glycol with water molecules or the use of salt solutions such as sodium acetate to neutralize the negative charge of phosphatidylserine on the surface of exosomes, thereby reducing the solubility and dispersibility of exosomes and forming aggregates. Finally, the exosome particles can be harvested by centrifugation [61, 65, 73]. The most commonly used commercial polymer precipitation product for exosome isolation is the ExoQuick proprietary polymer from System Biosciences (Santa Clara, CA, USA) [83]. Compared with other methods, the exosome RNA isolated by the ExoQuick precipitation method has the largest amount and the highest purity [84]. However, exosomes isolated by this method are susceptible to contamination by free proteins such as viral particles, immunoglobulins, and lipoproteins [74]. Higher purity EVs generally have a higher ratio of particles (mainly vesicle components) to proteins (vesicular content proteins and other contaminating proteins), so this ratio can be directly used to reflect the purity of EV samples [85].

There are a large number of proteins and receptors on the exosome membrane. Immunoaffinity capture uses the specific binding of receptors and their ligands or the immunoaffinity between these proteins and their specific antibodies to separate exosomes [63]. Surface receptors such as CD9, CD63, and CD81 are considered biomarkers of exosomes and are often used to extract exosomes by the immunoaffinity capture method [86]. The main techniques developed based on this principle are the enzyme-linked immunosorbent assay in microtiter plate format and affinity capture based on immunomagnetic beads [60]. Given the specificity of the interaction between the antigens on the exosome membrane and the corresponding antibodies, theoretically, specific antibodies for tumor-associated antigens can separate tumor-derived exosomes from exosomes of nontumor origin. The problem is that these antigens may be overexpressed in tumor cells, but their concomitant expression on normal cells can lead to immune capture of exosomes in the patient’s plasma, making it impossible to guarantee that all isolated exosomes are derived from tumor cells [87]. Nonetheless, as a rare exception, epitopes of melanoma-associated antigens are expressed only on melanoma cells and not on normal cells. Researchers have used immunoaffinity capture technology to separate melanoma-derived exosomes from other normal cell-derived exosomes in patient plasma [87]. In addition, isolation using immunoaffinity capture technology was shown to be better than centrifugation and density-based methods for the isolation of human colon cancer cell line-derived exosomes [78]. However, selectivity can also result in concomitantly lower yields compared to procedures that rely on isolation by physical properties, as some markers may not be presented or recognized on all vesicles within a given class [77].

Microfluidic-based exosome isolation is a high-throughput method that uses a microfluidic device to isolate exosomes based on several characteristics including size, density, and immunoaffinity. In addition, there are many novel isolation methods in the microfluidic platform such as nanowire trapping, acoustics, lateral displacement, and viscoelastic flow [59, 88]. Compared with other existing exosome isolation methods, microfluidic methods are faster, less expensive, require fewer samples and reagents, can potentially isolate exosomes of specific cell origin, and retain most microfluidically isolated exosomes in their native form [80]. Davies et al. [89] isolated exosomes from whole blood via a microfluidic platform using pressure- or electrophoresis-driven, in situ photopatterned porous polymer monoliths as filter membranes. Microfluidic devices offer several advantages for utilizing exosomes as diagnostic tools, such as low cost, reliability, real-time diagnosis, and the ability to process microvolume liquid samples such as saliva, breast milk, blood, and urine. However, its sample capacity is too small, which is a significant disadvantage [81].

Cur, a phytochemical with anticancer effects, can inhibit the transcription of the miR-21, thereby inducing PI3K/protein kinase B/phosphatase and tensin homolog (PTEN), nuclear factor kappa B (NF-$\kappa$B), and programmed cell death protein 4 signaling pathways to affect apoptosis, migration, cell proliferation, drug resistance, and stemness, and ultimately exert inhibitory effects on cancer cells [114]. However, the poor water solubility, poor stability, and low bioavailability of cur limit its use as a drug [115]. After loading cur into exosomes derived from milk and intestinal epithelial cells, they were both taken up by undifferentiated and differentiated intestinal Caco-2 cells, and the antiproliferative activity of cur was increased by 34% after incorporation into milk exosomes and by 26% in intestinal epithelial cell-derived exosomes [116]. In addition, the solubility of cur encapsulated in exosomes was significantly better than that in the free state. Exosomes can improve cur stability under extreme conditions such as high temperature, low or high pH, and enzyme treatment [115]. These properties make exosome-encapsulated cur a potentially effective oral anticancer drug [115].

Taxanes are a class of cytotoxic diterpenoids commonly used to treat solid malignancies [117]. The representative drug is PTX, which is used for the treatment of prostate, bladder, breast, ovarian, lung, and esophageal cancers [117]. Similar to cur, PTX is also poorly water-soluble, limiting its clinical application. The current solution is to bind PTX to albumin in a reversible non-covalent manner, increasing its solubility. This formulation is superior to isotoxic doses of standard PTX with a significantly lower incidence of toxicity [118]. Paradoxically, however, randomized controlled trials have shown that hematologic and non-hematologic toxicities, including peripheral neuropathy, are more prevalent with albumin-bound PTX than with conventional solvent-based PTX [119]. Another way to increase the water solubility of PTX is to dissolve it in a 50:50 mixture of Cremophor EL (CrEL) and ethanol. However, CrEL is not an inert carrier; it can exert a series of biological effects such as severe allergic reactions, hyperlipidemia, dyslipoproteinemia, red blood cell aggregation, and peripheral neuropathy [120]. Exosomes have also been used to load PTX [19]. Through exosomes secreted by M1 macrophages, PTX is delivered to tumor tissues. Studies have found that exosomes act as carriers to deliver higher amounts of PTX to tumor sites while activating the NF-$\kappa$B pathway, which significantly enhances the antitumor effect [121]. Penetrating the BBB is a significant challenge in delivering anticancer drugs to brain tumors. By contrast, loading PTX into exosomes secreted by bEnd.3 murine immortalized brain endothelial cell line can significantly increase the cytotoxicity against U-87 MG glioblastoma cells, suggesting that exosomes with specific properties have the potential to deliver drugs through the BBB [113]. Similarly, Zhu et al. [122] demonstrated that c(RGDyk) peptide-modified human embryonic stem cell-derived exosomes loaded with PTX have excellent glioblastoma-targeting ability.

Cisplatin is the first and most widely used metal-based chemotherapy drug. It is often used to treat solid tumors such as head and neck, stomach, lung, cervical, ovarian, bladder, and testicular cancers [123]. However, side effects and drug resistance limit its use in cancer therapy [123]. Zhou et al. [124] found that cisplatin could be delivered by milk-derived exosomes via clathrin-independent endocytosis and could avoid endosome capture and thus avoid drug resistance. Furthermore, exosomes secreted by M1 macrophages were shown to synergistically enhance the therapeutic effect of cisplatin in a mouse Lewis lung cancer model [125].

Doxorubicin (DOX) is a highly potent anticancer chemotherapeutic drug approved for the treatment of various human cancers [126]. However, its fatal cardiotoxicity limits its clinical application [127]. Encapsulation of DOX into liposomes can alter biodistribution, improve pharmacokinetics, and reduce toxicity [126]. On the downside, prolonged circulation of liposomes in the blood circulation may lead to adverse reactions in patients, including skin toxicity [128]. It has been reported that exosomes can serve as a suitable carrier of DOX. DOX was loaded into exosomes from the A33${}^{+}$ LIM1215 CRC cell line, which could form complexes with A33 antibody-modified magnetic iron oxide nanoparticles (US) (A33Ab-US-Exo/Dox). The complexes can specifically target A33${}^{+}$ colon cancer cells without causing significant cardiotoxicity [129]. The study by Wei et al. [130] also demonstrated that compared with free DOX, DOX loaded in exosomes had a much higher IC${}_{50}$ in the myocardial H9C2 cell line, indicating that exosomes as a carrier can significantly attenuate the cardiotoxicity of DOX. Similar to PTX, DOX can also penetrate the BBB via exosomes and thus be delivered to glioma cells (GMs). Notably, DOX-loaded autologous GM-derived exosomes inhibit the proliferation of parental GMs more than heterologous GMs, suggesting that autologous exosomes may be more favorable for drug targeting [131]. On the other hand, by labeling exosomes with targeting ligands, DOX can be delivered to specific sites. Covalently modified exosomes with the mucin 1 aptamer can be used to deliver DOX for the treatment of colon adenocarcinoma [132]. DOX is loaded into exosomes modified with the $\alpha$v integrin ligand iRGD polypeptide. Both in vitro and in vivo experiments have shown that the iRGD-exosome drug can specifically target $\alpha$v integrin${}^{+}$ breast cancer cells [133].

Exosomes can also deliver other types of therapeutic drugs. For example, HEK-293T cells were used to stably express a fusion protein of uracil phosphoribosyltransferase and cytosine deaminase by transfection, and exosomes loaded with the fusion protein were isolated, which could convert 5-fluorocytosine prodrugs into active form to exert anticancer effects [146]. In addition to proteins and mRNAs, antisense oligonucleotides, long non-coding RNAs (lncRNAs), shRNAs, circular RNAs (circRNAs), and aptamers have also been introduced to exosome-mediated cancer therapy [147]. Zheng et al. [148] found that exosome-delivered lncRNA PTEN pseudogene 1 could regulate PTEN levels by competing with miR-17, thereby inhibiting bladder cancer progression. Similarly, circRNA BTG anti-proliferation factor 2-loaded exosomes from RBP-J-overexpressing macrophages were shown to inhibit the proliferation and invasion of glioma cells via the miR-25-3p/PTEN pathway [149].

Specific cell-derived exosomes exhibit a natural tropism for the cancer microenvironment, which is critical for improving the targeting properties of nanomedicines [150]. Macrophage-derived exosomes, for example, can target cancer cells through their surface-expressed LFA-1 protein and ICAM-1 that is overexpressed in most cancer cells [97]. Similarly, exosomes from T cell lineages are taken up by myeloid cells, whereas mature DC-derived exosomes are efficiently internalized by activated T cells [151].

The artificial modification of exosome surface ligands can also promote the development of receptor-mediated tissue targeting [6, 18]. Chimeric antigen receptor engineered T cells (CAR-T)-derived exosomes not only retain the targeting specificity of CAR-T cells but also can significantly reduce the side effects caused by CAR-based adoptive immunotherapy due to their cell-free nature and biological characteristics, thus replacing CAR-T cells as direct attackers for cancer therapy [152]. In addition, exosomes modified with polyethylene glycol and a somatostatin receptor 2 (SSTR2) monoclonal antibody enable targeted delivery of the histone deacetylase inhibitor romidepsin (FK228) to neuroendocrine carcinomas overexpressing SSTR2 cells, showing potency and low toxicity [153].

The production of exosomes is limited, and is far from meeting the requirements for large-scale production. Methods such as ultracentrifugation or filtration are laboratory-scale-based methods for isolating exosomes and cannot meet the needs of clinical applications [16]. Measures to increase exosome production include bioreactors, three-dimensional cultures, and microfluidic devices [147]. Watson et al. [154] developed a hollow fiber bioreactor-based method for the large-scale production of high-purity bioactive EVs, which can increase the yield by 40 fold compared to traditional methods. However, the EV population obtained by this method is more diverse, which raises the requirements for subsequent purification [154]. The other method is a bioreactor culture system based on S/XF microcarriers, which increases the yield by 5-fold compared to conventional static systems, likely due to the aggregate formation and lower oxygen levels [155]. In addition, the study also confirmed that hypoxia could promote the secretion of exosomes by cancer cells, which may be mediated by hypoxia inducible factor alpha [156]. Likewise, stressors such as cytostatics increase exosome secretion from tumor cells [157]. It is worth noting that when increasing exosome secretion from cells by these physical or chemical factors, it should be carefully evaluated whether it will adversely affect exosome safety and therapeutic efficacy [16]. As another method to improve exosome yield, three-dimensional culture based on biocompatible collagen hydrogels can significantly improve the yield and collection efficiency of exosomes, and even modulate the biological function of exosomes [158]. The microfluidic co-flow system of the viscoelastic sample fluid and Newtonian sheath fluid reported by Liu et al. [159] can achieve 96% and 91% separation efficiency and recovery rate of exosomes below 200 nm, respectively. However, the current microfluidic platform is mainly suitable for small-volume sample separation. Expanding the device with a multichannel format may be one of the solutions for its future as a separation method for large-scale production [159].

Many studies have shown that exosomes from the same parental cell may have different molecular compositions [16]. In addition, the functions of a large part of the components are still unclear, such as some non-coding RNAs, lipids, and proteins. Incorporating these unknown biologically active molecules may bring certain risks [12]. Exosome cargoes such as miRNAs and lncRNAs, which have been widely studied, may be involved in the process of inducing or promoting tumor formation, which is undoubtedly dangerous for patients [6]. This is of particular concern in exosomes secreted by tumor cells, although they have the advantage of potentially targeting specific tumors. In the clinical setting, MSCs-derived exosomes are not only safe and reliable but also have the benefit of high yield [160, 161]. As drug delivery vehicles, the characterization of exosomes is crucial. Methods to characterize EVs with single particles such as transmission electron microscopy, flow cytometry, and nanoparticle tracking analysis have been reported, which have a driving role in understanding exosomes [162]. In addition, standardized quality management practices are critical to address the problems caused by exosome heterogeneity.

Studies have shown that intravenously infused exosomal drugs are rapidly cleared by the reticuloendothelial system in the spleen and liver, resulting in reduced drug accumulation at the target site. This rapid clearance may pose toxicity concerns for chemotherapeutic drugs [17]. The half-life of exosomes in vivo can be prolonged by blocking receptors that mediate exosome uptake. Watson et al. [154] showed that blockade of the scavenger class A family of receptors significantly reduces EV uptake by macrophages/monocytes. Modification of the exosome membrane is another way to prolong its half-life in vivo. CD47 is one of the exosomal proteins, and in the case of its high expression, exosomes can escape the phagocytosis of circulating monocytes and macrophages. Their half-life is significantly prolonged [163]. This may be achieved through the interaction of CD47 with signal regulatory protein alpha on phagocytes [164]. In addition, surface modification of exosomes with polyethylene glycol significantly inhibits the nonspecific uptake of DCs [161].

The low efficiency of current drug-loading methods largely limits their use, and the destruction of exosome integrity and drug stability is a common disadvantage of traditional methods. Some methods have been proposed to solve these problems to some extent. For example, exosome membrane protein CD9 is fused with RNA-binding protein human antigen R (HuR), and the resulting fusion protein recruits target RNA to exosomes through RNA-HuR recognition, markedly improving exosome nucleic acid-loading efficiency [165]. In addition, the chemical reagent-based method developed by Zhang et al. [166], which directly transfects miRNAs into exosomes, has proven convenient and efficient and can manipulate specific miRNAs in exosomes. On the other hand, many factors affect the loading efficiency of drugs such as the source of exosomes, the ratio of exosomes and drugs, and the properties of drugs [167]. These factors should be fully considered before the formulation design of exosomes, and the optimal parameters can be screened experimentally.