Aptamer-Conjugated Extracellular Nanovesicles for Targeted Drug Delivery

Extracellular nanovesicles (ENV) are cell-derived lipid bilayer–enclosed entities with size ranging from 30 to 300 nm (1). ENVs are secreted by many cell types and have been identified in diverse body fluids (2). They are specialized in intercellular communications facilitating transfer of cargo proteins and nucleic acids (3). In tumor, growing evidence indicates that ENVs can regulate tumor immune response, initiate formation of premetastatic niche, determine organotropic metastasis, contribute to chemotherapeutic resistance, and enable liquid biopsy for cancer diagnostics (4). Moreover, ENVs have been exploited as drug vehicles for drug delivery (5). Compared with micelles, liposomes, and polymeric nanoparticles, ENVs as a natural delivery system can evade phagocytosis, have extended blood half-life, and exhibit optimal biocompatibility without potential long-term safety issues (6). ENVs can fuse with the cell membrane and deliver drugs directly into cytoplasm. By evading the engulfment by lysosomes, ENVs remarkably enhance delivery efficiency of vulnerable molecules (5, 7, 8). In addition, the small size of ENVs facilitates their extravasation, translocation through physical barriers, and passage through extracellular matrix (9). Although ENV-based drug delivery is promising, two major challenges exist. First, the yield of ENVs is low (10). In general, ENVs can be collected from either cultured cell supernatant or autoplasma of patients. Given the secretion rate of ENVs ranges from 50 to 150 per cell per hour, the mass production of ENVs for potential clinical translation would be costly (11, 12). On the other hand, the prevalent ENV isolation techniques, such as ultracentrifugation, fail to generate highly pure ENVs (13), let alone the low isolation efficiency. Second, the preparation procedure for current ENV-based targeted drug delivery systems is tedious and labor-intensive. The typical approach is to secrete targeting peptide- or protein-grafted ENVs by transfecting the donor cells with corresponding recombinant virus or plasmid (14, 15). Of note, these transfection approaches have been criticized for inefficiency, potential toxicity, and insertional mutagenesis (16). The “click chemistry” has been used to conjugate targeting moieties onto EV surface in one-step (17). However, the conjugation procedure is time consuming, and the reaction conditions must be well controlled to avoid EV disruption and aggregation (18, 19). Hence, direct surface modification of EV is nontrivial either. In brief, for cancer treatment, nevertheless, simple approaches for large-scale production of tumor-targeting ENVs are desirable.

Here, we report a method for preparing aptamer-grafted ENVs for paclitaxel loading (Fig. 1). A nucleolin-targeting aptamer AS1411 was covalently conjugated to cholesterol-poly(ethylene glycol) (cholesterol-PEG) followed by anchoring the compounds onto living mouse dendritic cell (DC) membrane. Afterward, cells were extruded by passing through microconstrictions to obtain exosome-mimetic ENVs, and paclitaxel was subsequently loaded into ENVs with sonication. Using this aptamer-grafted ENVs, we investigated the ENVs biodistribution and whether paclitaxel can be more effectively delivered both in vitro and in vivo.

Cells were cultured in a humidified atmosphere of 5% CO₂ at 37°C. The MDA-MB-231 cells (ATCC, received in May 2016, passed fluorescence testing for Mycoplasma contamination on August 1, 2017, no more than 50 passages) were maintained in DMEM supplied with 10% FBS. Immature mouse DCs were isolated from bone marrow (20). BALB/c mouse was euthanized. After removal of the femurs, both ends of femurs were trimmed. The contents of marrow were flushed with 2 mL of HBSS with a needle, and bone marrow cells were washed with HBSS thrice. Approximately 2 × 10⁶ cells were cultured in RPMI1640 containing 10% FBS and 0.2 μg rmGM-CSF. Surface expression of CD11c (Santa Cruz Biotechnology, sc-23951) was analyzed.

FITC-tagged stearoyl-PEG (C18-PEG₂₀₀₀), distearoyl phosphethanolamine-PEG (DSPE-PEG₂₀₀₀), and cholesterol-PEG (chol-PEG₂₀₀₀) were dissolved in ethanol at 1 mmol/L. A total of 3 nmol of each lipid probe was added to 10⁷ DCs in 250 μL of PBS, respectively. The samples were incubated at 4°C for 5 minutes followed by washing thrice. Cells were fixed, stained with DAPI, rinsed with PBS, and finally resuspended in 500 μL of PBS. Cell suspension (20 μL) was added onto slips for imaging.

The 100 nmol of disulphide-tagged AS1411 aptamers (Sangon) were cleaved by TCEP solution on ice for 30 minutes (21), followed by incubation with 500 nmol of chol-PEG₂₀₀₀-maleimide (NanoCS) in HEPES buffer (Sigma-Aldrich) overnight at 4°C. The excess chol-PEG₂₀₀₀-maleimide was removed by filtration (MWCO 5000) at 10,000 × g for 15 minutes. The purified FITC-AS1411-PEG₂₀₀₀-chol was examined by matrix-assisted laser desorption/ionization—time-of-flight—mass spectrometry (MALDI-TOF-MS). A total of 1–5 nmol of FITC-AS1411-PEG₂₀₀₀-chol was added to 10⁷ DCs in 250 μL of PBS, respectively. Cells were incubated at 4°C for 5 minutes followed by washing with PBS thrice. The fluorescence intensity was measured using an Infinite M200.

A total of 10⁷ AS1411-PEG₂₀₀₀-chol labeled DCs in 1-mL PBS were continuously filtered with 10-μm and 5-μm track-etched membrane (Whatman) at 2,000 rpm for 5 times, and each filtration took 5 minutes. The supernatant was centrifuged at 300 × g for 5 minutes followed by 16,500 × g for 20 minutes. The supernatant was filtered using 0.22-μm filter. The collection of ENVs spontaneously secreted from DCs follows a general protocol. In brief, DCs were cultured in serum-free medium (SFM) for 48 hours. The medium was centrifuged at 16,500 × g for 20 minutes. Afterwards, the medium was filtered using a 0.22-μm filter followed by ultracentrifugation at 100,000 × g and 4°C for 2 hours. The ENV pellets were suspended in 500 μL of PBS.

Five microliters of AS1411-ENV samples were seeded onto substrate and fixed. The morphology of AS1411-ENVs was confirmed under Zeiss FESEM. Five microliters of AS1411-ENV sample was placed on 400 mesh grids and incubated for 3 minutes at room temperature. Excess samples were blotted with filter paper and stained with 1% uranyl acetate for 1 minute. Samples were then examined in a FEI Tecnai TEM. For cryo-TEM, 5 μL of AS1411-ENV sample was applied to a 200 mesh grids blotted for 1 second with FEI Vitrobot before plunging into liquid ethane. Samples were visualized in FEI Tecnai F20 TEM. The number of AS1411-ENVs was measured using Nanosight LM10 (Malvern). The characterization of ENVs follows the same protocol.

After lysis with RIPA buffer, protein amount in AS1411-ENVs and ENVs were determined using Micro BCA Protein Assay (Pierce). Protein samples were analyzed using acrylamide gels, and then transferred onto polyvinylidene difluoride membranes. The protein blot was blocked for 1 hour at room temperature with 5% nonfat dry milk in PBS/0.05% Tween and incubated overnight at 4°C with Santa Cruz Biotechnology antibodies against Annexin II (sc-28385), TSG101 (sc-7964), HSC70 (sc-7298), CD9 (sc-13118), CD59 (sc-133170), and CD55 (sc-51733). Afterward, secondary antibodies were incubated for 1 hour at room temperature. Samples were washed with PBS/0.05% Tween 20 for 10 minutes thrice. Blots were developed with chemiluminescence.

Ten microliters of AS1411-ENV or ENV samples were seeded onto slips and incubated at 4°C for 1 hour. The slip was blocked with BSA (10 mg/mL) for 20 minutes followed by PBS rinsing. A complementary DNA probe of AS1411 aptamer (Cy5-5′-CCA CCA CAA CCA CC-3′) and a control random DNA probe (Cy5-5′-TGC TGT GAG TGA ACC TGC TGT GTT GA-3′) in 50 μL of solution (pH 8.4, 50 mmol/L KCl and 1.5 mmol/L MgCl₂) was incubated with AS1411-ENVs or ENVs in a humidified chamber at 37°C for 2 hours (22–24). After hybridization, the slip was washed with PBS for fluorescence analysis.

Paclitaxel and nanovesicles were mixed in number ratio of 10⁶. The mixture was sonicated using a Model 505 Sonic Dismembrator with 0.25-inch tip with the following settings: 20% amplitude, 6 cycles of 30 seconds on/off for 3 minutes with a 2-minute cooling period between each cycle. After sonication, the mixture was incubated at 37°C for 1 hour. Alternatively, paclitaxel and nanovesicles were mixed in the same ratio, stirred, and incubated for 1 hour at room temperature. Excess free drug was removed from paclitaxel-loaded AS1411-ENVs or ENVs with a Sephadex G25 column. The amount of loaded paclitaxel in respective group was measured by high-performance liquid chromatography (HPLC). To measure paclitaxel release, freshly prepared paclitaxel-loaded AS1411-ENVs or ENV were placed in a 300K MWCO float-A-lyzer G2 device (Spectrum Laboratories). The device was then placed in PBS at room temperature with stirring. Samples were taken at time points and analyzed by HPLC.

The cytotoxicity of paclitaxel, paclitaxel-loaded ENVs, and paclitaxel-loaded AS1411-ENVs in MDA-MB-231 cells was evaluated using MTT assay (Thermo Fisher Scientific). Approximately 4,000 MDA-MB-231 cells were treated with paclitaxel in various concentrations, ranging from 0 to 3,300 nmol/L. After 24 hours, 20 μL of MTT solution (5 mg/mL) was added to each well and incubated for 4 hours. Medium was discarded, and formazan precipitate was dissolved in 10% SDS in DMSO containing 0.6% acetic acid. The microplates were shaken in the dark for 30 minutes, and absorbance at 570 nm was measured. The IC₅₀ value was determined from the dose–response curves.

MDA-MB-231 cells in 6 groups were, respectively, treated with PBS, ENVs without paclitaxel, AS1411-ENVs without paclitaxel, 200 nmol/L bare paclitaxel, paclitaxel-loaded ENVs, and paclitaxel-loaded AS1411-ENVs for 24 hours. A total of 5 × 10⁵ harvested MDA-MB-231 cells in 100 μL 1 × binding buffer (BD Biosciences) were incubated with 10 μL PI and 5 μL of Annexin V-FITC for 30 minutes at room temperature. Each sample was analyzed by flow cytometry. With the same experimental setup, cells in 6 groups were stained with PI only and photographed with high-content imaging system (Molecular Devices).

Approximately 2 × 10⁶ MDA-MB-231 cells in 50-μL PBS mixed with 50 μL of Matrigel were inoculated subcutaneously to the flanks of BALB/c mice (∼18–22 g, 6 weeks), and allowed to grow to a tumor size approximately 100 mm³. The mice were then randomly divided into 6 groups. Drug was intravenously administrated every 2 days (7.5 mg of paclitaxel-equivalent per kg of body weight per dose) for 3 weeks. Mice were euthanized to harvest tumors. To study pharmacokinetics and biodistribution of paclitaxel, 0.1–0.3 g tissue samples were collected at 6 time points after treatment. The analysis of paclitaxel by HPLC follows the published protocol (25). All animal experiments were approved by and performed in accordance with guidelines from the Institutional Animal Care and Use Committee (IACUC) of the Model Animal Research Center of the Second Affiliated Hospital of Southeast University (Nanjing, Jiangsu, China).

Results are presented as mean ± SD. Statistical comparisons were performed by paired Student t test or ANOVA.

To efficiently immobilize lipid-conjugated aptamers onto cell membrane for subsequent generation of aptamer-grafted ENVs, we first determined the lipid molecule type and optimal dose. PEGylated monoacyl lipid, diacyl lipid, and cholesterol are frequently used (26, 27). Hence, we chose FITC-tagged C18-PEG₂₀₀₀, DSPE-PEG₂₀₀₀, and cholesterol-PEG₂₀₀₀ for optimization. The three lipids can spontaneously insert into cellular lipid bilayer, but show differential labeling effects (Fig. 2A). The average fluorescence intensity of cells labeled with chol-PEG₂₀₀₀ was higher than that of the others (P < 0.01, Fig. 2B). There was no difference in fluorescence intensity between C18-PEG₂₀₀₀ and DSPE-PEG₂₀₀₀. Because C18-PEG₂₀₀₀ and DSPE-PEG₂₀₀₀ can readily assemble forming micelles or liposomes due to the long hydrophobic tail(s) and hydrophilic head, the self-assembly impairs cell labeling effect. In contrast, chol-PEG₂₀₀₀ can alleviate that to a certain extent as cholesterol itself is amphiphilic and relatively rigid. In addition, cholesterol also can increase the rigidity and stability of liposomes or EVs by enhancing the hydrophobic–hydrophobic interactions in lipid bilayers (28).

Because the stable retention of lipid-conjugated aptamer on cell surface favors generation of AS1411-ENVs, the retention of three lipids over time was evaluated with the corresponding decay of fluorescence at 4°C and 37°C, respectively. At 4°C, the fluorescence intensity of cells labeled with C18-PEG₂₀₀₀, DSPE-PEG₂₀₀₀, and chol-PEG₂₀₀₀, respectively, decreased to 93.2%, 97.71%, and 95.74%, indicating excellent retention over 100 minutes at low temperature (Fig. 2C). In contrast, at 37°C, the fluorescence intensity of cells labeled with C18-PEG₂₀₀₀, DSPE-PEG₂₀₀₀, and chol-PEG₂₀₀₀, respectively, gradually decreased to 48.63%, 56.98%, and 65.04% over the same period (Fig. 2C). We did not further measure the fluorescence intensities beyond 100 minutes as the extrusion of labeled cells for generating adequate ENVs only takes less than 30 minutes. Considering chol-PEG₂₀₀₀ demonstrated a superior labeling effect and an excellent retention at low temperature, we chose chol-PEG₂₀₀₀ for the following studies.

Next, we optimized the dose of chol-PEG₂₀₀₀ for cell membrane labeling. The insufficient label with cholesterol-conjugated aptamers causes unsaturated aptamer immobilization on cell surface. On the contrary, overlabeling might trigger plasma membrane bubbling and destroy the integrity of the lipid bilayer. The fluorescence intensity significantly increased with the dose of chol-PEG₂₀₀₀ ranging from 1 to 4 nmol (P < 0.05; Fig. 2D). To identify the labeling efficiency of chol-PEG₂₀₀₀, we compared the fluorescence intensity of FITC-tagged chol-PEG₂₀₀₀ at various doses before and after labeling (Supplementary Fig. S1A). The labeling efficiency (fraction of fluorescence intensity difference over initial FITC-tagged chol-PEG₂₀₀₀) of chol-PEG₂₀₀₀ ranging from 1 to 5 nmol (1–5 μL of 1 mmol/L chol-PEG₂₀₀₀ added into 250 μL of cell suspension, Supplementary Fig. S1B) was 15.33%, 20.03%, 20.05%, 22.09%, and 18.73%, respectively (Supplementary Fig. S1C). Therefore, we chose 4 nmol of chol-PEG₂₀₀₀ for labeling of approximately 10⁷ cells, and accordingly we determined that approximately 5.3 × 10⁷ chol-PEG₂₀₀₀ molecules were incorporated into an individual DC.

AS1411-PEG₂₀₀₀-chol was synthesized and confirmed by MALDI-TOF-MS (Fig. 2E). Excess free lipids were removed, and then the amount of AS1411-PEG₂₀₀₀-chol was determined by UV-visible spectroscopy. Meanwhile, DCs were isolated and expanded in vitro in 5 days (Supplementary Fig. S2A). Following the optimized protocol of cell labeling, we identified FITC-tagged AS1411-PEG₂₀₀₀-chol can be grafted onto DC membrane (Fig. 2F). After labeling, we loaded approximately 10⁷ DC cells (purity: approximately 90%, characterized with CD11c, Supplementary Fig. S2b) into an extruder for generating AS1411-ENVs (29). The morphology of partial AS1411-ENV was observed with a super-resolution microscopy (Nikon) in SIM mode (Fig. 2G). The average diameter of AS1411-ENV measured by microscopy ranged from 110 to 310 nm. Nanosight revealed the size distribution of AS1411-ENVs ranged from 50 to 270 nm with peak concentration at 103 nm. In contrast, in the control group, the diameter of ENVs spontaneously secreted from unlabeled DCs was 30–240 nm with peak concentration at 98 nm. In addition, approximately 8.7 × 10¹⁰ AS1411-ENVs were prepared by extrusion of approximately 10⁷ cells in 25 minutes. Under the assumption that the whole-cell membrane would participate in forming ENVs during extrusion, the production efficiency of AS1411-ENVs using our protocol was nearly 30%, and there is still room for improvement. In comparison, only 3.4 × 10¹⁰ ENVs were harvested from approximately 6.75 × 10⁷ cells cultured in SFM for 48 hours. The generation efficiency of nanovesicles by extrusion was approximately 17-fold higher and 115-fold faster than that of spontaneous secretion from cells. Under electron microscopy, there was no significant difference in morphology between AS1411-ENVs and ENVs (Fig. 2H). Both displayed a typical saucer-shaped morphology. The cryo-TEM and cryo-SEM images also demonstrate that both were spherical and enclosed by a lipid bilayer.

We also directly incubated bare ENVs with FITC-tagged AS1411-PEG₂₀₀₀-chol and determined the labeling efficiency. Given we used 4 nmol to label approximately 10⁷ cells, following the same ratio of lipid amount to total membrane surface area, we labeled approximately 7 × 10¹⁰ bare ENVs with 2 nmol/L of AS1411-PEG₂₀₀₀-chol in 250-μL suspension at 4°C for 5 minutes. After removal of free aptamer–conjugated lipids with ultracentrifugation, we recovered approximately 3 × 10¹⁰ AS1411-ENVs. The size at peak concentration was 105 nm (Supplementary Fig. S3A). Moreover, we found the fluorescence intensity of AS1411-ENVs prepared by extrusion (∼5 × 10⁸ in 10 μL) was higher than that of equivalent AS1411-ENVs prepared by direct labeling (P < 0.05, Supplementary Fig. S3B). We speculated that under the same experimental setting labeling ENVs with nanoscale lipids might be inefficient than microscale cells' labeling. The operation of ENV labeling is not comparable with cell labeling either. The additional purification with ultracentrifugation not only requires at least 2-hour processing time but also further decreases the final output. Therefore, direct ENV labeling with targeting ligands works despite low efficiency.

Cargo proteins extracted from approximately 2 × 10⁸ AS1411-ENVs and ENVs were analyzed by Western blot analysis. Three cytosolic proteins, including Annexin II, TSG101, and HSC70 and three surface proteins, including CD59, CD9, and CD55, were identified in both AS1411-ENVs and ENVs (Fig. 2I), indicating that these proteins can be enclosed into AS1411-ENVs during extrusion of cells. The CD55 and CD59 protect extracellular vesicle (EV) from attack of the complement system, making them stable in the blood (30, 31). Tetraspanin CD9 facilitates direct membrane fusion with the target cell and contents-delivery directly into cytosol (32). In addition, we chose immature DCs as they are devoid of lymphocyte stimulatory molecules such as MHC II, CD80 or CD86 (33). The stability of AS1411-ENVs and ENVs were examined by monitoring changes in size at respective peak concentration. The cryopreserved AS1411-ENVs and ENVs with approximately 100 nm in diameter did not show significant aggregation or degradation in PBS at −80°C over 6 months (Fig. 2J), indicating that AS1411-ENVs could be prepared in large-scale, cryopreserved in batches for at least 6 months, and then thawed for use.

We further used Cy5-conjugated hybridization probe (h-probe) of AS1411 and random probe as a control (c-probe) to reconfirm the existence of AS1411 on conjugated ENVs. The fluorescence intensities demonstrated only the dot blots of AS1411-ENVs incubated with h-probes showed strong red fluorescence signals (Fig. 2K). In comparison, due to a minute amount of residual probes, weak fluorescence signals were detected in the rest three groups. Next, approximately 8 × 10⁵ AS1411-ENVs and ENVs were incubated with MDA-MB-231 cells expressing nucleolin on cell membrane, respectively (34). We identified that AS1411-ENVs efficiently bound MDA-MB-231 cells in only 10 minutes (Fig. 2L and M). In contrast, equal amount of PKH67-labeled ENVs barely fused with MDA-MB-231 cells at the same time point. After 30 minutes, a few ENVs attached and fused with cell membrane, while significant amount of AS1411-ENVs were observed in the experimental group (P < 0.01). The findings indicate that AS1411-ENVs can rapidly recognize and bind onto cancer cell surface nucleolin, and thus would facilitate targeted drug delivery.

Paclitaxel was loaded into AS1411-ENVs and ENVs (in absolute number ratio of paclitaxel/vesicle: 10⁶), respectively, using either sonication or simple incubation at room temperature. The average paclitaxel-loading efficiency with sonication was 21.33% ± 1.53% in AS1411-ENVs and 26.4% ± 2.67% in ENVs. In comparison, paclitaxel loading efficiency with simple mixing, stirring and incubation was 5.06% ± 1.12% in AS1411-ENVs and 7.73% ± 2.38% in ENVs, respectively. During sonication, the membrane microviscosity significantly decreases, allowing drug molecules to enter vesicle, and the membrane integrity can be restored immediately after sonication. It does not significantly affect the membrane-bound proteins or the lipid contents of the ENVs (35). Of note, the negatively charged aptamers on the outer surface might affect encapsulation of negatively charged paclitaxel, causing slightly lower loading efficiency in comparison with that of ENVs. As the loading efficiency of sonication was much higher than that of incubation, we used sonication for paclitaxel loading in the following studies. After paclitaxel loading, the whole-size distribution showed a slight rightward shift, and the size of AS1411-ENVs and ENVs at peak concentration increased to 111 and 109 nm (black curves), respectively (Fig. 3A). In addition, after loading of paclitaxel, the mean zeta potential of AS1411-ENVs decreased from −23.7 mv to −25.6 mv, and that of ENVs decreased from −11.8 mv to −16.1 mv (Fig. 3B). The three cytosolic proteins including Annexin II, TSG101, and HSC70 were barely detectable in approximately 2 × 10⁸ AS1411-ENVs and ENVs after paclitaxel loading, respectively. These syntenins might be released from vesicles during sonication (Fig. 3C).

Moreover, we investigated the release kinetics of paclitaxel from AS1411-ENVs and ENVs at 37°C, respectively (Fig. 3D). Both paclitaxel-loaded nanovesicles showed burst release within the first 1 hour, and then displayed a sustained release profile thereafter. At 24-hour time point, paclitaxel-loaded AS1411-ENVs and ENVs released approximately 71.47% and 86.29% paclitaxel, respectively. The slow release of paclitaxel from nanovesicles is favorable for drug administration. Especially in targeted drug delivery, more drugs in vectors thus can be effectively delivered to the lesion. The ability to deliver paclitaxel into target cells in vitro was studied by identifying the IC₅₀ value of paclitaxel-loaded AS1411-ENVs, paclitaxel-loaded ENVs, and bare paclitaxel, respectively, after cell treatment for 24 hours, and we determined the respective IC₅₀ value of paclitaxel in the above three groups as 14.24, 35.08, and 212.69 nmol/L (Fig. 3E). The treatment efficacy of paclitaxel-loaded AS1411-ENVs and paclitaxel-loaded ENVs exhibited a 15-fold and 6-fold enhancement compared with bare paclitaxel in vitro. Of note, AS1411 as therapeutic effects, and AS1411 for acute myelogenous leukemia is being evaluated in phase II clinical trial (36). On the basis of generated AS1411-ENV amount and AS1411 grafting amount, we deduced that each individual AS1411-ENV bears approximately 2,000 AS1411 molecules. The amount is adequate for tumor targeting in vivo (37), but has very limited therapeutic effects. The dose of AS1411 in tumor therapy in vivo is 10 mg/kg/d or even higher, whereas the amount of AS1411 for targeting is less than 4% of the therapeutic dose. We speculated the significant decrease of the IC₅₀ value in group of AS1411-ENVs was mainly attributed to high delivery efficiency of paclitaxel to cells through the targeted ENVs.

We used 200 nmol/L paclitaxel-loaded AS1411-ENVs, paclitaxel-loaded ENVs, bare paclitaxel, and three negative controls to treat MDA-MB-231 cells, respectively. In groups of paclitaxel-loaded AS1411-ENVs and paclitaxel-loaded ENVs, approximately 90% of cells were effectively blocked in G₂–M phase, whereas in bare paclitaxel group, approximately 60% of cells had been retained in G₂–M phase (Fig. 3F and G). The paclitaxel-loaded AS1411-ENVs and paclitaxel-loaded ENVs demonstrated higher therapeutic efficacy in comparison with bare drugs as paclitaxel-loaded nanovesicles could efficiently deliver drugs to target cells by membrane fusion and AS1411-mediated internalization. In contrast, bare negatively charged paclitaxel molecules rely on passive diffusion for crossing negatively charged cell membrane. The entry mode bypasses the endosomal–lysosomal pathway (38), and thus both AS1411-ENVs and ENVs as delivery vehicles can circumvent the need for endosomal escape strategies. The high-content screening also confirmed significantly more dead cells (stained in red) in groups of paclitaxel-loaded AS1411-ENVs and paclitaxel-loaded ENVs in comparison with that of bare paclitaxel group and three negative control groups (Fig. 3H). Of note, after paclitaxel treatment, dead cells detached from flask surface during staining and rinsing, and thus fewer cells were left in the three experimental groups.

Next, we investigate the potential of paclitaxel-loaded AS1411-ENVs and paclitaxel-loaded ENVs in drug delivery using the MDA-MB-231 xenograft BALB/c mouse model. Tissue samples (heart, liver, spleen, lung, kidney, and tumor) were collected, and paclitaxel concentration in each organ was analyzed with HPLC. At 30 minutes postadministration, paclitaxel underwent rapid distribution to the main organs in each group (Fig. 4A). In bare paclitaxel group, paclitaxel easily diffused into other organs but less paclitaxel accumulated in tumors. In the next 12 hours, bare paclitaxel was readily cleared from main organs and tumors. In groups of paclitaxel-loaded AS1411-ENVs and paclitaxel-loaded ENVs, paclitaxel concentration in tumors was significantly higher than that in bare paclitaxel group at all time points, indicating encapsulation of paclitaxel into AS1411-ENVs and ENVs can achieve in vivo redistribution of paclitaxel. The clearance rate of paclitaxel-loaded AS1411-ENVs and paclitaxel-loaded ENVs were also slower than that of bare paclitaxel. We noted that both paclitaxel-loaded nanovesicles show relatively high concentration of paclitaxel in liver and spleen at the 12-hour time point. A previous study also reported that EVs displayed a predominant localization of intravenously administered EVs in the spleen and the liver (39). In general, the intravenous injection of nanoparticles with size over 50 nm results in intensive recognition by microphages that take nanoparticles to the liver, meanwhile 150–250 nm nanoparticles also frequently accumulate in the spleen (40). In addition, paclitaxel-loaded AS1411-ENVs continued to accumulate in tumors and reached peak concentration at 12-hour time point, showing significantly different (P < 0.01) pharmacokinetics compared with the other two groups (Fig. 4A). The tumor volumes of mice treated with PBS (from 167.06 ± 12.20 to 1,041.59 ± 238.3 mm³), ENVs without paclitaxel (from 167.55 ± 12.36 to 1,059.67 ± 339.68 mm³), and AS1411-ENVs without paclitaxel (from 168.53 ± 18.04 to 1,110.32 ± 284.27 mm³) rapidly increased, and there was no significant difference among the three negative controls (Fig. 4B and C). The final tumor weight of mice treated with PBS, ENVs without paclitaxel, and AS1411-ENVs without paclitaxel were 0.94 ± 0.20 g, 0.96 ± 0.32 g, and 1.00 ± 0.26 g, respectively. Tumors that received bare paclitaxel showed a slight growth inhibition due to low bioavailability, and thus tumor volume and weight increased from 168.96 ± 16.56 mm³ to 283.28 ± 90.65 mm³ by the end of treatment. paclitaxel-loaded ENVs inhibit the growth of tumor, and the volume decreased from 169.7 ± 17.01 mm³ to 97.99 ± 30.08 mm³. Paclitaxel-loaded AS1411-ENVs significantly restrained growth of grafted tumors, and the tumor volume decreased from 169.58 ± 18.04 mm³ to 35.34 ± 3.38 mm³ (P < 0.05), which could be attributed to the active targeting effect in tumor treatment (Fig. 4B and C). At the end of treatment, the relative tumor volume of each group was 6.19 ± 1.01 (without paclitaxel), 6.24 ± 1.49 (ENVs without paclitaxel), 6.58 ± 1.4 (AS1411-ENVs without paclitaxel), 1.66 ± 0.43 (bare paclitaxel), 0.57 ± 0.15 (paclitaxel-loaded ENVs), and 0.21 ± 0.04 (paclitaxel-loaded AS1411-ENVs), respectively. Average tumor weight of paclitaxel-treated three groups was 0.18 ± 0.089 g (bare paclitaxel), 0.05 ± 0.016 g (paclitaxel-loaded ENVs), and 0.02 ± 0.002 g (paclitaxel-loaded AS1411-ENVs), respectively (Fig. 4D). The in vivo therapeutic efficacy of paclitaxel-loaded AS1411-ENVs was improved approximately 9- and 3-fold compared with that of bare paclitaxel and paclitaxel-loaded ENVs, respectively (P < 0.05). There was no significant difference in mouse body weight during the 3-week administration (Fig. 4E). Furthermore, tumor proliferation and the histologic structures of tumors and organs from all groups were analyzed (Fig. 4F; Supplementary Fig. S4). Compared with other formulations, paclitaxel-loaded AS1411-ENVs caused remarkable tumor tissue damage and reduced the percentage of proliferating Ki67-positive tumor cells, indicating the enhanced treatment efficacy. We also measured the systemic toxicity. Bone marrow suppression is the major dose-limiting toxicity of paclitaxel, and dose-/schedule-dependent neutropenia is the typical hematologic toxicity. Over 10 days after intravenous administration, a routine blood test was performed to exam leukocytes counting and function enzymes of liver and kidney. The mean counts of leukocytes, erythrocytes, and platelets were relatively lower in bare paclitaxel-treated mice; meanwhile, the average level of five major indexes of kidney and liver function was also relatively higher (Fig. 4G). However, no significant difference was observed among 6 groups, and their levels were well within the normal physiologic range. The findings also indicated paclitaxel-loaded AS1411-ENVs or paclitaxel-loaded ENVs overcame the systemic toxicity. The tumor-bearing mice could tolerate higher dose of paclitaxel-encapsulated in targeting ENVs without the occurrence of severe systemic toxicities. The encapsulation would increase the chemotherapeutic effects while minimizing exposure to normal and healthy cells.

The field of EV-based drug delivery has greatly expanded over the last few years, and many studies have eloquently demonstrated that EV can function as therapeutic nanocarriers for delivering a variety of cargos, including siRNAs, miRNAs, proteins, and chemotherapeutics to particular tissue (41–44). Compared with existing delivery systems, the outstanding advantage of EVs as delivery vehicles is their favorable lipid and surface protein composition. Currently, cells such as epithelial cells and model cancer cell lines have been used as donor cells for harvesting derived EVs. However, immunogenicity and biosafety concerns arise for in vivo applications of these donor cells derived EVs. Particularly, cancer cell line–derived EVs as drug delivery vehicles might possess tumorigenic potential due to incomplete depletion of cargos and/or inherent tumor-associated proteins imbedded on EV membrane. In comparison, DCs can be harvested from a patient's bone marrow or peripheral blood by standard protocols (45). They should thus be less immunogenic than any foreign delivery vehicle. The in vitro multiplication of DCs can constantly provide EVs, which is extremely important for multiple episodes of treatment. On the other hand, at present, only a small portion of published reports utilize a targeting strategy for direct delivery of the therapeutic cargos. The mainstream routine for harvesting derived EVs relies on expression of targeting protein or peptide on cell membrane after sophisticated gene transfection (8, 14, 33, 46). Alternatively, targeting ligands can be covalently immobilized onto membrane proteins via click chemistry. However, both technique routines are quite challenging. Moreover, the workload required to upscale the production of cells and the resulting yield of EVs is currently not suitable for clinical use.

The method we developed can easily create adequate targeting ligand–grafted ENVs in 1 hour for drug loading. Our approach has multiple advantages. First, both targeting ligands (lipid-conjugated aptamers or palmitoylated antibodies) and donor cells (immune cells, stem cells, or others) can be arbitrarily geared to meet actual need. Theoretically, lipidated ligands targeting estrogen-related receptor beta 2 (ERRB2), prostate-specific antigen (PSA), carbonic anhydrase IX (CA IX), or other targets frequently used in chimeric antigen receptor (CAR) T-cell therapy could be produced in large-scale and grafted onto autologous immune cells for targeted therapy of different cancers. The whole procedure would be very flexible. On the contrary, the bioengineered cells through gene transfection only express designated protein or peptide. To construct cells expressing another specific membrane protein, the preparation of viral or plasmid vectors, transfection, and selection have to be performed once again. Second, donor cells potentially can be obtained from patient's autologous cells to generate ENVs for drug loading. Genetic engineering is not necessary. Therefore, it is much safe than cell-based immunotherapy, such as CAR T-cell therapy. Although CAR T-cell therapy was recently approved, the efficacy, toxicity, potential mutagenesis and easy-to-operation are still big concerns (47, 48). Third, we demonstrated generation efficiency and rate of ENVs via extrusion of cells is 17- and 115-fold higher than that of ENVs spontaneously released by donor cells. Cells harvested from a T75 flask can generate approximately 8 × 10¹⁰ ENVs via extrusion, satisfying a single dose, and the production can easily be scaled up. The relatively low cost and the ease of operation would benefit the clinical implementation. Fourth, using our method, donor cells can be safely, efficiently, and homogenously labeled with lipid-conjugated ligands in assigned ratio within 5 minutes, ensuring derived individual ENVs grafted with targeting ligands. Cell labeling with aptamer-conjugated cholesterol only needs to determine the amount of cells and lipidated aptamers. In contrast, neither immobilization of targeting ligands onto EV membrane via click chemistry nor in fusion of EVs with ligands grafted liposomes can be easily and precisely performed at nanoscale (49). Altogether, the extrusion of targeting ligands grafted cells for generating ligands functionalized ENVs is a rapid, easy, and economic approach to prepare sufficient drug delivery vehicles. The enhanced in vivo therapeutic efficacy and low systemic toxicity make this approach conductive to potential clinical translation in the future.

No potential conflicts of interest were disclosed.

Conception and design: Y. Wan, C. Zhu, Q. Zheng, G. Wang, J. Tong, Y. Fang, Y. Xia, G. Cheng, X. He, S.-Y. Zheng

Development of methodology: Y. Wan, L. Wang, C. Zhu, Q. Zheng, G. Wang, J. Tong, Y. Fang, Y. Xia, G. Cheng, X. He, S.-Y. Zheng

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Wan, L. Wang, C. Zhu, Q. Zheng, G. Wang, J. Tong, Y. Fang, Y. Xia, G. Cheng, X. He, S.-Y. Zheng

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Wan, L. Wang, C. Zhu, Q. Zheng, G. Wang, J. Tong, Y. Fang, Y. Xia, G. Cheng, X. He, S.-Y. Zheng

Writing, review, and/or revision of the manuscript: Y. Wan, L. Wang, C. Zhu, Q. Zheng, G. Wang, J. Tong, Y. Fang, Y. Xia, G. Cheng, X. He, S.-Y. Zheng

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Wan, C. Zhu, Q. Zheng, G. Wang, J. Tong, Y. Fang, Y. Xia, G. Cheng, X. He, S.-Y. Zheng

Study supervision: C. Zhu, Q. Zheng, G. Wang, J. Tong, Y. Fang, Y. Xia, G. Cheng, X. He, S.-Y. Zheng

S. Zheng thanks the Penn State Materials Research Institute, the Huck Institute of Life Sciences, the Penn State Microscopy and Cytometry Facility. This work was supported by Nanjing Science and Technology Development Foundation Number 201605031 (to Q. Zheng), Jiangsu Provincial Medical Youth Talent Award QNRC2016054 (to L. Wang), Nanjing Medical Science and Technology Development Foundation-Youth Program Number ZDX16008 (to X. He), Natural Science Foundation of Jiangsu Province BK20170134 (to C. Zhu), and National Cancer Institute of the NIH under award number DP2CA174508 (to S. Zheng).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.