Milk-derived exosomes as a promising vehicle for oral delivery of hydrophilic biomacromolecule drugs

Abstract

Exosomes, as promising vehicles, have been widely used in the research of oral drug delivery, but the generally low drug loading efficiency of exosomes seriously limits its application and transformation. In this study, we systematically investigated the effects of drug loading methods and physicochemical properties (lipophilicity and molecular weight) on drug loading efficiency of milk-derived exosomes to explore the most appropriate loading conditions. Our finding revealed that the drug loading efficiency of exosomes was closely related to the drug loading method, drug lipophilicity, drug molecular weight and exosome/drug proportions. Of note, we demonstrated the universality that hydrophilic biomacromolecule drugs were the most appropriate loading drugs for milk-derived exosomes, which was attributed to the efficient loading capacity and sustained release behavior. Furthermore, milk-derived exosomes could significantly improve the transepithelial transport and oral bioavailability of model hydrophilic biomacromolecule drugs (octreotide, exendin-4 and salmon calcitonin). Collectively, our results suggested that the encapsulation of hydrophilic biomacromolecule drugs might be the most promising direction for milk exosomes as oral drug delivery vehicles.

Graphical abstract

1. Introduction

Exosomes (EXOs) are extracellular vehicles (EVs), with size of 30–150 nm, secreted by various cells and can be isolated from many biological fluids, including blood, urine and breast milk [1], [2], [3]. EXOs are considered as promising drug delivery vehicles due to their natural nano-size, low immunogenicity and intrinsic ability to penetrate biological barriers [4,5]. Thereinto, bovine milk EXOs exhibit excellent oral biocompatibility, stable source, and are easier to achieve large-scale production, which have attracted much attention in the field of oral drug delivery [6], [7], [8].

Some evidences have demonstrated the unique advantage of bovine milk EXOs in transporting various drugs across the gastrointestinal barrier [9], [10], [11]. It was reported that paclitaxel-loaded bovine milk EXOs could improve the solubility of paclitaxel, enhance its oral absorption, thus solve the problem that oral administration of paclitaxel cannot reach the therapeutic dose due to its poor water solubility [12]. Our group previously revealed that bovine milk EXOs loaded with biological macromolecule insulin could avoid its degradation in the severe physiological environment of gastrointestinal tract. Meanwhile, oral delivery of insulin-loaded bovine milk EXOs elicited a more superior and more sustained hypoglycemic effect when compared with subcutaneously injection of insulin on type I diabetes model [13]. Such evidences suggested the enormous potential of bovine EXOs for oral drug delivery.

However, due to the limited loading space, the generally low drug loading efficiency of EXOs could not be neglected [14,15], and it greatly limited the application and transformation of EXOs as drug carriers. In order to improve the drug loading efficiency of EXOs, a variety of drug loading methods have been developed, including incubation, saponin, sonication and freeze/thaw and so on [16,17]. Disappointedly, no specific loading technique could properly encapsulate all drugs at present. Therefore, it is necessary to select appropriate methods to match the physicochemical characteristics of drugs, and the compatibility between different loading methods and different drugs needs to be studied. According to the current researches, the loading efficiency of milk-derived EXOs for hydrophobic small molecules such as curcumin, anthocyanin and paclitaxel varied from10 to 40% [18], but the hydrophilic large molecule like insulin was more than 50% [13]. Such difference reminds us that the drug loading efficiency of milk EXOs is closely related to the physicochemical properties of drugs [19,20]. Nevertheless, more detailed relationships and effects of physicochemical properties of drugs on loading efficiency of bovine milk EXOs remain to be elucidated. Moreover, on the basis of drug loading, the controlled drug release is also crucial for oral drug delivery [21]. Only the sustained and slow release can realize the protective effect of EXOs on drugs and the ability of EXOs to deliver drugs across the intestinal epithelial barrier [13,22]. However, whether milk EXOs can achieve controlled drug release after loading different drugs has not been reported yet. In conclusion, a better understanding of the loading and controlled release capacity of bovine milk EXOs for different drugs during its application and transformation in the field of oral drug delivery is necessary.

In this work, four common drug loading methods of EXOs, incubation, saponin, sonication and freeze/thaw, were used to study the loading efficiency and release behavior of drugs with different lipophilicity and molecular weight. Our results demonstrated that the lipophilicity of drugs was the main factor affecting the loading efficiency of EXOs, the loading efficiency of hydrophilic drugs was significantly higher than hydrophobic drugs. Meanwhile, the larger molecular weight, the higher loading efficiency when using saponin and sonication to loaded hydrophilic drugs, while the molecular weight of hydrophobic drugs barely affected their loading efficiencies. On the contrary, molecular weight had a great influence on drug release. In short, milk EXOs have high loading efficiency and continuous and slow drug release loaded with hydrophilic macromolecules. Notably, the loading and release of multiple hydrophilic macromolecules (especially biological macromolecular peptides) by EXOs were similar to the above results, which proved that the results were not accidental. Finally, we validated in vitro and in vivo that the oral absorption of hydrophilic macromolecules could be facilitated by loading in bovine milk EXOs. Therefore, milk-derived EXOs were promising vehicles for oral delivery of hydrophilic biomacromolecule drugs.

2. Materials and methods

2.1. Materials

5-Fluorouracil was purchased from Aladdin Biochemical Technology Co., Ltd. (Shanghai, China). Colchicine was obtained from Shanghai Yien Chemical Technology Co., Ltd. (Shanghai, China). Carbamazepine was obtained from National Institutes for Food and Drug Control. Celecoxib and paclitaxel were purchased from Dalian Meilun Biotechnology Co., Ltd. (Dalian, China). Glimepiride was bought from Shanghai Bide Medical Technology Co., Ltd. (Shanghai, China). Octreotide was purchased from Suzhou Tianma Pharmaceutical Group Tianji biopharmaceutical Co., Ltd. (Jiangsu, China). Liraglutide, exendin-4 and salmon calcitonin were obtained from Jiante Biological Medicine Co., Ltd. (Chengdu China). Saponin was purchased from Shanghai Maikelin Biochemical Technology Co., Ltd. (Shanghai, China). 1,1ʹ-Dio-ctadecyl-3,3,3ʹ,3ʹ-tetramethylindocarbocyanine perchlorate (DiI) was purchased from Invitrogen (Carlsbad, CA, USA). Fluorescein isothiocyanate (FITC) was obtained from J&K Scientific Ltd. (Beijing, China). All other chemical reagents in the study were of analytical grade or above.

2.2. Isolation of milk-derived EXOs

Defatted bovine milk (Inner Mongolia Yili Industrial Group Co., Ltd.) was purchased from a local supermarket. The defatted milk and acetic acid were mixed (100:1, v/v) for 5 min at room temperature followed by centrifugation at 12,000×g for 15 min at 4 °C. The Supernatant was filtered with a 0.22-µm membrane and obtained whey. The whey was ultracentrifuged at 210,000×g for 70 min at 4 °C (Beckman Coulter, SW40Ti rotor, Brea, CA, USA) [23]. The EXO pellets were resuspended in PBS and ultracentrifuged again. After the wash, the pellet was resuspended in PBS, and centrifugated at 10,000×g for 5 min at 4 °C to remove the residual precipitates. The total protein concentration was measured by BCA kit. Then, EXOs (<6 mg/ml) were stored at −80 °C until use.

2.3. Characterization of EXOs: DLS, TEM, and WB

Dynamic light scattering (DLS) size and zeta potential of EXOs and EXO@drug were measured by Zetasizer Nano ZS90 (Malvern Instruments Ltd., UK). The morphology of EXOs was observed using transmission electron microscope (TEM), and the EXO samples were negatively stained with phosphotungstic acid (2%) for 3 min before observation. Western blot (WB) technique was applied to detected the maker proteins CD63, TSG 101 and the contaminant casein of milk derived EXOs. The exosomal proteins were subjected to SDS-PAGE and electro-transferred to a PVDF membrane. The membrane was incubated with specific primary rabbit polyclonal antibodies overnight at 4 °C, and secondary horseradish peroxidase (HRP)-conjugated goat anti-rabbit Ig-HRP at 37 °C, 2 h. Then proteins bands were visualized by chemiluminescent substrate and imaged on a Bio-Rad ChemiDoc MP imager.

2.4. In vitro stability study of EXOs

To test the colloidal stability, the EXOs were suspended in simulated gastric fluid (SGF, pH 2.0), simulated intestinal fluid (SIF, pH 6.8) and phosphate buffered saline (PBS, pH 7.4), respectively, and incubated in a shaker at 60 rpm for 8 h at 37 °C. Then the sizes of EXOs were measured at predetermined times.

The structural stability of EXOs was investigated by fluorescence resonance energy transfer (FRET) assay. First, DiO and DiI were selected as a FRET pair, and DiO labeled exosomes (DiO-EXOs), DiI labeled exosomes (DiO-EXOs) and DiO/DiI co-labeled exosomes (DiO/DiI-EXOs) were prepared [13]. Next, EXOs were incubated in SGF, SIF, and PBS at 37 °C for 8 h. The fluorescence intensity of dual-loaded and single-loaded EXOs was detected at determined time points by Varioskan Flash Multimode Reader (Thermo Fisher Scientific, San Jose, CA, USA) with an excitation wavelength of 460 nm and the emission spectrum was recorded from 490 to 620 nm. The energy transfer efficiency (E) was calculated by the following formula:

$$E=1-F_{\text{DA}}/F_{\mathrm{D}}$$

where F_D and F_DA represent the DiO fluorescence intensity of the donor alone (DiO-loaded EXOs) and with the presence of the acceptor (DiO/DiI-loaded EXOs), respectively.

2.5. Preparation of drug-loaded EXOs

Four approaches for drug incorporation into EXOs were evaluated: incubation, saponin, sonication and freeze/thaw cycles. For incubation, EXOs and drug solution were mixed in 1 ml PBS to incubated 15 min at 37 °C. In case of a saponin, a mixture of drug and EXOs was supplemented with 0.2% saponin and co-incubated for 15 min at 37 °C [24]. For sonication, EXOs and drug were mixed followed by sonication (100 W, 3 min, 2 s/2 s on/off) for 3 cycles with a two minutes cooling period between each cycle [25]. After sonication, drug-loaded EXOs solution was incubated at 37 °C for 60 min to allow for recovery of the exosomal membrane. As for freeze/thaw cycles, the drug solution was added to EXOs as described above, then rapidly freezed in liquid nitrogen, and thawed at 37 °C. The freeze/thaw cycle was repeated three times [17].

In order to improve loading efficiency as much as possible, different loading ratios were used for hydrophilic and hydrophobic drugs. Specifically, EXOs (200 µg, content of proteins)/drug (1 mg) proportion was 1:5 (w/w) in the loading process of four hydrophilic drugs, while the proportion was 1:1 (EXOs 200 µg/drug 200 µg) for four hydrophobic drugs and hydrophobic drug solution (dimethyl sulfoxide, DMSO) was mixed with the EXO dispersion by keeping the final solvent concentration ≤2%. The unloaded hydrophilic drug was removed by ultrafiltration (MWCO: 100 kDa), and for hydrophobic drug, for hydrophobic drugs, low-speed centrifugation at 10,000×g for 10 min was required before ultrafiltration. The content of loaded drug was quantified by high performance liquid chromatography (HPLC), and the loading efficiency (LD) of EXO@drug was calculated according to the following equation:

$$\text{Loading}\text{efficiency}(\%)=\frac{W_{\text{drug}}}{W\text{proteins}}\times 100\%$$

where W_drug is the weight of loaded drug, W_proteins is the total weight of exosomal proteins.

2.6. Drug release study

To evaluate drug release behavior, drug-loaded EXOs freshly prepared by different methods were transferred into dialysis tubes (MWCO: 100 kDa), which were immersed in SGF (pH 1.2, 2 h) and in SIF (pH 6.8, 6 h) under sink conditions at 37 °C with stirring. Additionally, 0.2% Tween 80 was also added to the SGF and SIF to maintain the sink conditions when measured the hydrophobic drug release [12]. Samples were taken at time points from inside the dialysis tube and were analyzed by HPLC as described above. The amount of drug released from EXOs was expressed as a percentage of total drug and plotted as a function of time.

2.7. Transepithelial transport study

Before the experiment, we evaluated the toxicity of EXOs treated with sonication or saponin by Alamar Blue assay. In brief, Caco-2 cells were seeded in 96-well plates and cultured for 3 d. After removal of DMEM and washed by PBS, cells were incubated with blank EXOs treated with sonication or saponin and with exosomal protein concentrations ranged from 0 to 400 µg/ml for 6 h. Then, cells were washed twice with cold PBS to separate extracellular EXOs. Finally, Alamar Blue DMEM solution (10 µg/ml) was added into each well (150 µl/well) for 1 h and the absorption of each well was measured at Ex: 490 nm/ Em: 570 nm via a Varioskan Flash Multimode Reader.

In the study of transepithelial transport, Caco-2 were seeded into Tanswell® insert and cultured for 21 d (TEER ≈ 300 Ω cm²). Firstly, the cells were equilibrated with the blank medium at 37 °C. Then, the apical medium was replaced by the 200 µl medium with EXO@FITC-OCT, EXO@FITC-Lira, EXO@FITC-E4 or EXO@FITC-sCT and FITC-labeled corresponding free drugs were used as control group. FITC-labeled peptides were synthesized as previously reported to facilitate detection [26,27]. At predetermined time intervals (0.5, 1, 2, 3, 4, and 6 h), 50 µl samples were withdrawn from the basolateral chamber, and the equal volume fresh medium was supplemented. After broken the sample solution with DMSO, the fluorescence intensity was detected by a microplate reader and the apparent permeability coefficient (P_app) was calculated using the following equation:

$$P\text{app}=\frac{d_{\mathrm{Q}}}{d_{\mathrm{t}}}\times \frac{1}{A\times C_0}$$

where d_Q/d_t is the rate of EXOs diffusion from chamber A to chamber B; C₀ is the initial concentration of EXOs; A is the membrane area (cm²).

2.8. In vivo pharmacokinetic studies

To evaluate the blood drug level and bioavailability of drug-loaded EXOs, 200–250 g healthy male Sprague–Dawley rats were randomly divided into three groups (n = 6 per group). Prior to experiments, rats were fasted overnight but with free access to water. The following formulations were administered to rats individually: oral drug solution (OCT: 2.64 mg/kg; Lira: 5.00 mg/kg; E4: 0.5 mg/kg or sCT: 1.00 mg/kg), oral EXO@drug (OCT: 2.64 mg/kg; Lira: 5.00 mg/kg; E4: 0.5 mg/kg or sCT: 1.00 mg/kg), and SC injection of drug solution (OCT: 0.264 mg/kg; Lira: 0.20 mg/kg; E4: 0.05 mg/kg or sCT: 0.10 mg/kg). Blood samples were collected from the posterior orbital vein, centrifuged (3000 rpm, 3 min), and subsequently quantified. Among them, E4 plasma concentration was quantified using an ELISA kit (EK-070-94, Phoenix Pharmaceuticals Inc.), sCT was also detected using an ELISA kit (EK-014-09, Phoenix Pharmaceuticals Inc.); and the plasma concentration of OCT and Lira was measured by LC/MS. AUC of drug concentration and relative bioavailability (Fr%) were then calculated according to previous reports. All animals received care in compliance with relevant laws and guidelines outlined in the Guide for the Care and Use of Laboratory Animals. All procedures were approved by Sichuan University Animal Care and Use Committee.

2.9. Statistical analysis

Data were presented as mean ± standard deviation (SD). All statistical analyses were performed using two-tailed Student's t-test when two groups were compared, or one-way analysis of variance (ANOVA) with Tukey's post-hoc test when multiple groups were compared in SPSS 19.0 software. P < 0.05 was considered significant.

3. Results and discussion

3.1. Isolation and characterization of EXOs

In this study, EXOs were isolated and obtained by an established protocol combining acid precipitation and ultracentrifugation. DLS measurement showed that EXOs had an average size of 103.79 ± 0.79 nm with narrow polydispersity index (PDI) (Fig. 1A), and the zeta-potential was −12.8 mV. TEM showed spherical and cup-shaped EXOs with a diameter of approximately 100 nm (Fig. 1B). WB analysis demonstrated that the bands of exosomal marker proteins CD63 and TSG101 in the whey were shallow and casein content was high. However, the EXOs obtained from whey by ultracentrifugation were rich in typical EXO marker proteins, including membrane binding protein CD63 and cytoplasmic protein TSG 101, and the contaminant casein was extremely low (Fig. 1C). These results indicated that EXOs with high purity were successfully isolated from whey. Then, the stability of EXOs was examined, as illustrated in Fig. 1D, the size of EXOs did not significantly change in PBS (pH 7.4), simulated gastric fluid (SGF, pH 2.0) and simulated intestinal fluid (SIF, pH 6.8) during 8 h incubation (37 °C). In addition, the results of FRET (Figs. 1E and S1A and S1B) also showed obvious energy transfer effects when EXOs were co-labeled with DiO and DiI, but the efficiency of this energy transfer process in PBS, SGF and SIF did not change significantly within 8 h (Fig. 1F). Together, the above results indicated the successful isolation of EXOs from bovine milk, and the excellent morphological and structural stability of obtained EXOs in PBS, SGF and SIF.

Fig. 1.

Characterization of EXOs. (A) Size detected by DLS. (B) TEM images of EXOs. Scale bar: 200 nm. (C) WB of exosomal marker proteins and contaminant. (D) The size variation of EXOs in PBS, SGF and SIF during 8 h. Error bars represent SD (n = 3). (E) FRET assay in PBS. The arrow indicated the FRET signal. (F) The change of FRET efficiency in PBS, SGF and SIF during 8 h. Error bars represent SD (n = 3).

3.2. Compatibility of different loading methods to drugs with different properties

Four hydrophilic drugs (i.e., 5-fluorouracil, colchicine, octreotide and liraglutide) and four hydrophobic drugs (i.e., carbamazepine, celecoxib, glimepiride, and paclitaxel) were used as model drugs to study the adaptation of different loading methods to drugs with different properties. The compounds were chosen to represent a wide range of molecular weight with additional differences in lipophilicity, and the physicochemical properties of all 8 drugs evaluated are described in Table 1.

Table 1.

Structures and physicochemical properties of model drugs.

Drug	Structures	Mw (g/mol)	log Pb
5-Fluorouracil (5-FU)		130.08	−0.90
Colchicine (COL)		399.40	1.00
Octreotide (OCT)	D-F-[C-F-D-W-K-T-C]-T-OLa	1019.20	−0.49
Liraglutide (Lira)	H-A-E-G-T-F-T-S-D-V-S-S-Y-L-E-G-Q-A-A-K -[N-(1-oxohexadecyl)-L-γ-glutamyl]-E-F-I-A-W-L-V-R-G-R-Ga	3751.00	−3.40
Carbamazepine (CAR)		296.27	2.77
Celecoxib (CEL)		381.40	3.40
Glimepiride (GLI)		490.60	3.90
Paclitaxel (PTX)		853.90	3.00

^aAbbreviation for the amino acid sequence of a polypeptide.

^blog P: oil−water partition coefficient, the log P value is proportional to the lipophilicity.

3.2.1. Compatibility of different loading methods to hydrophilic drugs

First, we investigated the impacts of different encapsulating approaches on the loading efficiencies of hydrophilic drugs. As illustrated in Fig. 2A, among the four loading methods, the loading efficiency of 5-fluorouracil in EXOs (EXO@5-FU) by sonication was the highest (37.65%), followed by freeze/thaw (35.21%), and there was no significant difference between them. The lowest loading efficiency was 14% by incubation. The loading efficiencies of sonication and freeze/thaw were significantly higher than saponin and incubation, and the loading efficiencies of EXO@5-FU decreased successively: sonication ≈ freeze/thaw > saponin > incubation.

Fig. 2.

Loading efficiencies of EXOs for hydrophilic drugs obtained by four methods. (A) 5-Fluorouracil, (B) colchicine, (C) octreotide, (D) liraglutide. *P < 0.05, **P < 0.01, ***P < 0.001 versus incubation group. ^#P < 0.05, ^##P < 0.01, ^###P < 0.001 versus saponin group. ^$P < 0.05, ^$$P < 0.01, ^$$$P < 0.001 versus sonication group. Error bars represent SD (n = 3).

The loading efficiencies of EXOs for colchicine (EXO@COL) were slightly different from EXO@5-FU (Fig. 2B), the highest loading efficiency was achieved by saponin (33.38%), followed by sonication (28.04%), but there was no significant difference between them. In consistence with EXO@5-FU, the loading efficiency of EXO@COL obtained by the incubation was just 9.25%, which was significantly lower than the other three methods. The loading efficiencies of EXO@COL were decreased following the row: saponin ≈ sonication > freeze/thaw > incubation.

For OCT (Fig. 2C), the decreasing trend for loading efficiencies obtained by four methods was similar to EXO@COL, with saponin > sonication > freeze/thaw > incubation, and the loading efficiencies were between 64.77%−14.40%. There were significant differences among different methods.

As shown in Fig. 2D, sonication achieved the highest loading efficiency (62.14%), when liraglutide was loaded into EXOs (EXO@Lira), followed by saponin, and there was no significant difference between them. As always, the incubation had the lowest loading efficiency at 14.73%. The loading efficiencies of EXO@Lira decreased successively: sonication ≈ saponin > freeze/thaw ≈ incubation.

Taken together, for hydrophilic drugs, sonication and saponin generally achieved higher loading efficiencies, followed by freeze/thaw and incubation, namely sonication ≈ saponin > freeze/thaw > incubation. Therefore, from the perspective of loading efficiency, sonication and saponin are more suitable for the loading of hydrophilic drugs by milk-derived EXOs. This regularity was consistent with previous reports [17,24]. During sonication, the mechanical shear force from the sonicator probe weakens the membrane of the EXOs, facilitating the penetration of cargos into EXOs, which is conducive to the efficient loading of hydrophilic drugs [28]. Meanwhile, saponins are efficient membrane-penetrating agents which could form complexes with cholesterol in membrane and produce holes/pores, thus increasing the membrane permeability and loading efficiency [29]. On the contrary, it was difficult for hydrophilic drugs to cross the phospholipid bilayer of EXOs by incubation alone, so the loading efficiency was relatively lower than other methods. It's also worth mentioning that sonication is not suitable for the loading of drugs that are unstable after sonication, and the toxicity of saponin could not be neglected. These factors should be considered together in practical application. In addition, the particle size and zeta-potential of EXOs did not change significantly after loading with hydrophilic drugs (Fig. S2A–S2D).

3.2.2. Compatibility of different loading methods to hydrophobic drugs

Next, we investigated the impacts of different encapsulating approaches on the loading efficiencies of hydrophobic drugs. As illustrated in Fig. 3A, the encapsulation of hydrophobic small molecule carbamazepine in EXOs (EXO@CAR) by saponin was the highest (13.27%), and significantly higher than freeze/thawing, sonication and incubation. Different from the loading of hydrophilic drugs, the loading efficiency of EXO@CAR by sonication was the lowest. And the loading efficiencies of EXO@CAR were decreased following the order: saponin > freeze/thaw > incubation ≈ sonication.

Fig. 3.

Loading efficiencies of EXOs for hydrophobic drugs obtained by four methods. (A) Carbamazepine, (B) celecoxib, (C) glimepiride, (D) paclitaxel. * P < 0.05, ** P < 0.01, *** P < 0.001 versus incubation group. ^#P < 0.05, ^##P <0.01, ^###P < 0.001 versus saponin group. ^$P < 0.05, ^$$P < 0.01, ^$$$P < 0.001 versus sonication group. Error bars represent SD (n = 3).

For celecoxib (EXO@CEL, Fig. 3B), the loading efficiencies of the four methods ranged from high to low as saponin > incubation ≈ freeze-thaw > sonication. The loading efficiency of the saponin was 20.78%, while that of the sonication method was only 2.24%, which was significantly lower than the other three methods.

As Fig. 3C showed, the regularity of loading efficiencies of glimepiride (EXO@GLI) was consistent with EXO@CAR. The loading efficiencies ranged from 11.02% to 4.19% in the order of saponin > freeze/thaw > incubation ≈ sonication.

Same as above, saponin achieved the highest loading efficiency of 21.60% among the four drug loading methods When EXOs loaded paclitaxel (EXO@PTX), followed by incubation, sonication and freeze/thaw. There was no significant difference among the rest of three methods.

Obviously, the saponin significantly improved the loading efficiency of EXOs for hydrophobic drugs, which was similar with hydrophilic drugs. Disappointedly, in contrast to the loading regularity of hydrophilic drugs, the sonication was unable to achieve superior drug loading for hydrophobic drugs. In summary, the adaptation rules of the four drug loading methods for EXOs loaded with hydrophobic drugs were as follows: saponin >> freeze/thaw ≈ incubation ≈ sonication. As well, the particle size and zeta-potential of EXOs did not change significantly after loading with hydrophilic drugs (Fig. S3). Besides, gel electrophoresis and Coomassie experiments staining showed that the surface proteins of EXOs treated with four drug loading methods were similar to those of the original EXOs. This result demonstrated that the four drug loading techniques in this study did not significantly affect proteins expressed on EXO surfaces (Fig. S4).

3.3. Enhanced lipophilicity decreased drug loading efficiency of EXOs

We used milk EXOs to separately encapsulate four weak lipophilic drugs (Log P ≤ 1, Table 1) and four strong lipophilic drugs (log P > 1, Table 1) with various molecular weights. As indicated in Fig. 2, the loading efficiencies of EXOs were high for all four hydrophilic drugs with different molecular weights, with the highest loading efficiency ranging from 33.38% to 64.77%. However, the loading efficiencies of EXOs for hydrophobic drugs were significantly reduced, and the highest loading efficiency of the four drugs was only between 13.37% and 21.6% (Fig. 3). In conclusion, the lipophilicity of drugs is an important factor affecting the drug loading efficiency of EXOs, and the loading efficiencies of EXOs for hydrophilic drugs is significantly higher than that for hydrophobic drugs.

3.4. Effect of molecular weight of drugs on loading efficiency of EXOs

Since the lipophilicity could affect the drug loading efficiency of EXOs, the loading efficiencies of hydrophilic and hydrophobic drugs were respectively rearranged according to the molecular weight from small to large when discussing the influence of molecular weight on the loading efficiency.

3.4.1. Larger molecular weight facilitated drug loading for hydrophilic drugs

First, as shown in Fig. 4A–4D, the results showed that the loading efficiencies of EXOs for liraglutide (EXO@Lira) and octreotide (EXO@OCT) with larger molecular weight were significantly higher than that of the hydrophilic small molecule 5-fluorouracil (EXO@5-FU) and colchicine (EXO@COL) when loaded by saponin and sonication. (Fig. 4B and 4C). However, it is worth noting that no consistent regularity between molecular weight and loading efficiency was obtained when the other two methods were used, for example, there was little difference in the loading efficiency of drugs with different molecular weights for incubation, and the loading efficiency of EXOs for the macromolecular EXO@Lira and EXO@OCT had no significant differences compared with the small molecule EXO@5-FU (Fig. 4A). More unexpectedly, the loading efficiency of EXO@5-FU was significantly higher than EXO@COL, EXO@OCT and EXO@Lira when loaded by freeze/thaw (Fig. 4D).

Fig. 4.

Correlation between molecular weight and drug loading efficiency of EXOs. (A–D) was hydrophilic drugs loaded by four methods. (A) Incubation, (B) saponin, (C) sonication, (D) freeze/thaw. *P < 0.05, **P < 0.01, ***P < 0.001 versus EXO@5-FU. ^#P < 0.05, ^##P < 0.01, ^###P < 0.001 versus EXO@COL. ^$P < 0.05, ^$$P < 0.01, ^$$$P < 0.001 versus EXO@OCT. (E–H) was hydrophobic drugs loaded by four methods. (E) Incubation, (F) saponin, (G) sonication, (H) freeze/thaw. * P < 0.05, ** P < 0.01, *** P < 0.001 versus EXO@CAR. ^#P < 0.05, ^##P < 0.01, ^###P < 0.001 versus EXO@CEL. ^$P < 0.05, ^$$P < 0.01, ^$$$P < 0.001 versus EXO@GLI. The value in brackets on the abscissa is the molecular weight of the corresponding drug. Error bars represent SD (n = 3).

Based on the above results, for hydrophilic drug, molecular weight on the influence of the loading efficiency was associated with drug loading methods, the loading efficiency of EXOs increased with the increase of drug molecular weight when loaded by saponin or sonication, but the molecular weight of the drug had no significant effect on the loading efficiency of EXOs when the drug was loaded by incubation or freeze/thaw. Some studies indicate that the loading capacity is essentially determined by intermolecular interactions between drugs and carrier materials, including hydrophobic interaction, electrostatic interaction, hydrogen bonding, Pi–Pi stacking and van der Waals force [30]. The high loading capacity can be achieved with the enhanced interaction between drug and vehicle. Compared with small molecule drugs, there are more chemical groups in macromolecule drugs which are beneficial to produce hydrogen bonding, hydrophobic interaction and electrostatic interaction with EXOs. Therefore, we speculated that the superior loading of macromolecule drugs might be due to the stronger affinity between macromolecule drugs and milk-derived EXOs. In future work, the interaction between drug molecules and EXOs can be further studied to prove this conjecture.

3.4.2. No noteworthy correlation between molecular weight and drug loading efficiency of EXOs for hydrophobic drugs

Similarly, in hydrophobic drugs, the effect of drug molecular weight on the drug loading efficiency of EXOs was investigated. As demonstrated in Fig. 4E–4H, the loading efficiency of EXOs for hydrophobic drugs did not show an obvious trend of increasing or decreasing with the increase of drug molecular weight for all drug loading methods. As for hydrophobic drugs, there seemed to be no significant correlation between the drug loading efficiency of EXOs and the molecular weight of drugs, which may be caused by the following reasons. Firstly, restricted by the strong lipophilicity of hydrophobic drugs, the loading efficiencies of EXOs for hydrophobic drugs were generally low (<22%). Moreover, in clinical, most hydrophobic drugs are small chemical drugs (<1,000 Mw), while the molecular weights of hydrophilic macromolecules such as insulin commonly used in clinical can reach more than 3,000 Mw. Therefore, the range of molecular weight of hydrophobic model drugs (296.27–853.90 Mw) was much narrower than that of hydrophilic drugs (130.08–3,751 Mw) in our study. Taken together, generally low loading efficiency and relatively similar molecular weight of hydrophobic drugs might hardly reflect the effect of molecular weight on loading efficiency.

3.5. The loading efficiency of EXOs for hydrophobic drugs was mainly limited by their low solubility in water-based loading media

Previous studies have shown that the loading efficiency of EXOs for hydrophobic drugs is significantly lower than that of hydrophilic drugs. Remarkably, different EXO/drug proportion were used in the drug loading process for hydrophobic and hydrophilic drug, due to their solubility differences in water-based loading media. Specifically, EXO/drug proportions of 1:5 (µg EXOs protein content versus µg drug) were used for hydrophilic drugs, and 1:1 for hydrophobic drugs. Hence, we investigated the loading efficiency of EXOs for hydrophobic and hydrophilic drugs at a serious of same EXO/drug proportions to study the effect of solubility of drugs on loading efficiency. Among the four hydrophilic model drugs, 5-fluorouracil with the smallest molecular weight and liraglutide with the largest molecular weight were selected as representative drugs. Similarly, carbamazepine and paclitaxel were selected among the four lipophilic model drugs.

As shown in Fig. 5A, the loading efficiency of EXOs for 5-fluorouracil (EXO@5-FU) gradually increased with the increase of the EXO/drug proportions for all four methods, and the loading efficiency of the 1:5 group(EXO/drug) was significantly higher than the 5:1 and 1:1 group. Similarly, the regularity of loading efficiency for the macromolecular hydrophilic drug liraglutide (EXO@Lira) at different EXO/drug proportions was consistent with EXO@5-FU (Fig. 5B). However, the loading efficiency of EXOs for hydrophobic drugs showed a trend of increasing first and then decreasing with the increase of the EXO/drug proportions, and the maximum loading could be achieved basically when the proportion was 1:1 (Fig. 5C and 5D). The above results indicated that the low loading efficiency of EXOs for lipophilic drugs is not caused by low EXO/proportion, the low water solubility may be the main reason that limits the loading efficiency of EXOs for lipophilic drugs. In the process of EXO drug loading, drug molecules in external solution mainly enter the EXOs by passive diffusion [16,31], while different drug loading methods only create more favorable conditions for molecular diffusion. For hydrophilic drugs, high EXO/drug proportion forms strong concentration gradient outside the EXOs, which is conducive to the diffusion of drug molecules, thus achieving superior loading efficiency; but hydrophobic drugs tend to aggregate and precipitate in the aqueous solution, which leads to the decrease of loading efficiency, especially with high EXO/drug proportion.

Fig. 5.

Loading efficiency of EXOs at different EXO/drug proportions. (A) 5-Fluorouracil loaded by four methods. (B) Liraglutide loaded by four methods. (C) Carbamazepine loaded by four methods. (D) Paclitaxel loaded by four methods. *P < 0.05, **P < 0.01, ***P < 0.001 versus 5:1 group. ^#P < 0.05, ^##P < 0.01, ^###P < 0.001 versus 1:1 group. Error bars represent SD (n = 3).

3.6. Effects of drug properties on drug release from EXOs

On the basis of drug loading, the drug release of EXOs loaded with different drugs in SGF (2 h) and SIF (6 h) was further investigated. The release of hydrophilic drugs loaded by EXOs is shown in Fig. 6. Different from the limited effect of molecular weight on drug loading efficiency, molecular weight has a great influence on the drug release. The hydrophilic drug 5-fluorouracil and colchicine with small molecular weight showed obvious burst release, more than 50% of drug was released within half an hour, which may be caused by the adsorption of some drugs on the surface of EXOs. Among them, 5-fluorouracil and colchicine were released completely within 2 h (Fig. 6A) and 4 h (Fig. 6B), respectively. Surprisingly, with the increase of molecular weight of drugs, the rate of drug release from EXOs slowed down, and the burst release phenomenon disappeared. Octreotide was completely released within 6 h (Fig. 6C), and the cumulative release of liraglutide was only 55.43% in 8 h (Fig. 6D). In addition, different drug loading methods had no significant effect on the drug release rate. As illustrated in Fig. 6E, the release rate of the hydrophilic drugs loaded in EXOs decreased significantly with the increase of molecular weight within 8 h. Especially, EXOs loaded with macromolecules octreotide and liraglutide showed sustained and slow drug release.

Fig. 6.

Drug release of EXOs loaded with hydrophilic drugs in SGF and SIF. (A) Release of 5-fluorouracil loaded by different methods. (B) Release of colchicine loaded by different methods. (C) Release of octreotide loaded by different methods. (D) Release of liraglutide loaded by different methods. (E). Release rates of hydrophilic drugs with different molecular weights (all drugs were loaded by sonication, the number above the curve is the cumulative release of the corresponding drug within 8 h). *P < 0.05, **P < 0.01, ***P < 0.001 versus EXO@5-FU. ^#P < 0.05, ^##P < 0.01, ^###P < 0.001 versus EXO@COL. ^$P < 0.05, ^$$P < 0.01, ^$$$P < 0.001 versus EXO@OCT. Error bars represent SD (n = 3).

The drug release pattern of EXOs loaded with hydrophobic drugs was similar to hydrophilic drugs. Carbamazepine (Fig. 7A) and celecoxib (Fig. 7B) with smaller molecular weight both exhibited burst release, but in general, the release rate of lipophilic drugs from EXOs is slower than that of hydrophilic drugs. Although there was a sudden release of glimepiride within 0.5 h, its subsequent release rate slowed down, and the cumulative release was 88.17% within 8 h (Fig. 7C), while paclitaxel continued to release slowly within 8 h, and the cumulative release was 59.38%.

Fig. 7.

Drug release of EXOs loaded with hydrophobic drugs in SGF and SIF. (A) Release of carbamazepine loaded by different methods. (B) Release of celecoxib loaded by different methods. (C) Release of glimepiride loaded by different methods. (D) Release of paclitaxel loaded by different methods. (E). Release rates of hydrophobic drugs with different molecular weights (all drugs were loaded by sonication, the number above the curve is the cumulative release of the corresponding drug within 8 h). *P < 0.05, **P < 0.01, ***P < 0.001 versus EXO@CAR. ^#P < 0.05, ^##P < 0.01, ^###P < 0.001 versus EXO@CEL. ^$P < 0.05, ^$$P < 0.01, ^$$$P < 0.001 versus EXO@GLI. Error bars represent SD (n = 3).

Release rate of drugs from nanocarriers could be influenced by lots of aspects like hydrophilicity/hydrophobicity, distributions in nanocarriers, molecular weight and intermolecular forces. The molecular weight-dependent release profiles might be caused by multiple factors. On the one hand, benefit from lower molecular size and smaller fluid mechanics radius, small molecular drugs like 5-fluorouracil and colchicine were easier to penetrate through the phospholipid bilayer of EXOs, thus accelerating the releasing speed [32]. By contrast, high hydrodynamic volume of hydrophilic macromolecular drugs effectively limited the free diffusion, so they were released in gentler speed. On the other hand, some studies indicate that the weak intermolecular interactions between drug molecule and carrier material usually lead to a fast drug release, but the high loading capacity and slower release can be achieved with the enhanced interaction between drug and vehicle [31]. Therefore, we speculated that there may be stronger interaction between macromolecule drugs and EXOs, resulting in longer desorption time and thus slower drug release behavior. It's worth noting that release of macromolecules from EXOs is a complex process, which depends on desorption, diffusion, particle erosion or the combination of these factors and still not fully understood. We speculate that the initial release may be triggered by gradual desorption of macromolecules adsorbed on the EXO surface or partially embedded in the EXO membrane under sink conditions. In addition, although different drug loading methods did not change the particle size, zeta-potential and surface proteins of EXOs significantly, the influence on other properties of EXO membranes, such as permeability and compactness, is not clear at present, which is also the focus of our further research. Some studies have reported that sonication could reduce the microviscosity of EXO membrane, and saponins and freeze/thaw also could slightly damage EXO membrane. Therefore, we speculate that different drug loading methods may not significantly affect the integrity of EXOs, but loosen the phospholipid bilayer, which may be the reason for the slow release of macromolecular drugs from EXOs under sink conditions.

3.7. Universality of superior loading of EXOs for hydrophilic macromolecules

In order to investigate the universality of superior loading and sustained release of milk EXOs for hydrophilic macromolecules, salmon calcitonin (s-CT, Mw:3,431.85) and exendin-4 (E4, Mw: 3,369.76) were selected as model drugs to further study. As expected, the loading efficiencies of EXOs for s-CT and E4 were both high (LD > 50%, Fig. 8A and 8B), and the drug release in vitro was relatively slow. (Fig. 8C and 8D). Surprisingly, the efficient loading of hydrophilic macromolecules by milk EXOs is universal. Hence, four hydrophilic macromolecules: octreotide, liraglutide, exendin-4 and salmon calcitonin were selected as model drugs for subsequent studies on the oral delivery ability of EXOs.

Fig. 8.

Loading efficiency and drug release of exendin-4 (EXO@E4) and salmon calcitonin (EXO@sCT). (A) Loading efficiency of exendin-4 loaded by different methods. (B) Loading efficiency of salmon calcitonin loaded by different methods. *P < 0.05, **P < 0.01, ***P < 0.001 versus incubation group. ^#P < 0.05, ^## <0.01, ^###P < 0.001 versus saponin group. ^$P < 0.05, ^$$P < 0.01, ^$$$P < 0.001 versus sonication group. (C) Release of exendin-4 loaded by different methods. (D) Release of salmon calcitonin loaded by different methods. **P < 0.01, ***P < 0.001 saponin versus incubation, ^#P < 0.05, ^##P < 0.01, saponin versus sonication, ^$$P < 0.01, ^$$$P < 0.001 saponin versus freeze/thaw. Error bars represent SD (n = 3).

3.8. EXOs improved the transepithelial transport in vitro and oral bioavailability in vivo of drugs

The intestinal epithelial barrier is an important reason for limiting the oral absorption of drugs, especially biomacromolecules. Human colon carcinoma cell line (Caco-2) was selected as in vitro intestinal cell model to evaluate the absorption efficiency of EXOs. First, the results of resazurin showed that the drug-loaded EXOs prepared by saponin had strong cytotoxicity to Caco-2 cells within 6 h, while those prepared by sonication had no significant cytotoxicity (Fig. S5A and S5B). Therefore, the drug-loaded EXOs prepared by sonication were applied in subsequent experiments. In addition, for the convenience of quantification, polypeptides linked to FITC were synthesized for cell experiments. As shown in Fig. 9A–9D, apparent Papp value of drug-loaded EXOs was more than 2-fold enhanced compared with those of free drugs, and the transepithelial electrical resistance (TEER) value of cell monolayer showed no significant change after treatment (Fig. 9E–9H). Therefore, EXOs and drugs did not disrupt the integrity of the cell monolayer, excluding the paracellular route. EXOs significantly improved transepithelial transport of hydrophilic macromolecules, which was conducive to oral delivery.

Fig. 9.

The improvement of transepithelial transport and oral bioavailability of hydrophilic polypeptide macromolecules loaded by EXOs. (A–D) The P_app value of free drugs and drug-loaded EXOs. (A) FITC-octreotide. (B) FITC-liraglutide. (C) FITC-exendin-4. (D) FITC-salmon calcitonin (**P < 0.01). (E–H) The TEER value of the corresponding drugs. Error bars represent SD (n = 3). (I) Plasma octreotide level versus time profiles of SD rats. (J) Plasma liraglutide level versus time profiles of SD rats. (K) Plasma exendin-4 level versus time profiles of SD rats. (L) Plasma salmon calcitonin level versus time profiles of SD rats. *P < 0.05, **P < 0.01, ***P < 0.001 versus corresponding drug solution (p.o.) group. Error bars represent SD (n = 6).

To demonstrate the oral absorption efficacy of drug-loaded EXOs, we measured time-dependent plasma drug levels and their related pharmacokinetic parameters (Table 2) on healthy Sprague-Dawley rats. As shown in Fig. 9, the plasma levels of the drugs were significantly increased when the drug was encapsulated into EXOs. Among them, free octreotide was poorly absorbed after oral administration. After encapsulated by EXOs, the blood concentration increased significantly and peaked at 0.5 h (Fig. 9I), the oral relative bioavailability was 1.23% which was 5.59 times higher than that of the free octreotide (Table 2). Similarly, the oral relative bioavailability of exendin-4 increased from 0.76% to 2.84% after EXOs loading, a 3.74-fold improvement over the oral free exendin-4 (Fig. 9K and Table 2); and for salmon calcitonin, the oral relative bioavailability after EXOs loading was 2.23%, a 2.86-fold increase compared with the free salmon calcitonin (Fig. 9L and Table 2). Taken together, EXOs could significantly increase the plasma concentration and AUC of hydrophilic macromolecular drugs after oral administration, and thus improved their oral relative bioavailability. EXOs showed great potential in oral delivery of hydrophilic macromolecular drugs.

Table 2.

Pharmacokinetic parameters of plasma drugs in healthy rats.

Sample	Dose(mg/kg)	AUC0–6h (µg/l·h)	Fr%
OCT solution (s.c.)	0.264	312.41 ± 51.11	—
OCT solution (p.o.)	2.64	6.99 ± 2.66	0.22
EXO@OCT	2.64	38.48 ± 9.12	1.23
Lira solution (s.c.)	0.20	1254.70 ± 230.22	—
Lira solution (p.o.)	5.00	62.70 ± 4.69	0.20
EXO@Lira	5.00	314.35 ± 138.38	1.00
E4 solution (s.c.)	0.05	11.97±1.50	—
E4 solution (p.o.)	0.50	0.92±0.80	0.76
EXO@E4	0.50	3.40±1.50	2.84
sCT solution (s.c.)	0.10	25.94 ± 1.89	—
sCT solution (p.o.)	1.00	2.02 ± 0.51	0.78
EXO@sCT	1.00	5.79 ± 2.44	2.23

4. Conclusion

In this study, we systematically investigated the effects of drug loading methods and physicochemical properties of drugs on drug delivery performance of EXOs. Specifically, drug loading methods deeply impacted the loading efficiency. Thereinto, saponin could efficiently encapsulate various drugs with different properties, and sonication was more suitable for encapsulating hydrophilic drugs, while the loading efficiency of freeze/thaw and incubation was weak. Of note, the physicochemical properties of the drugs also greatly affected the loading capacity of EXOs. Hydrophilic drugs showed generally higher loading efficiency than hydrophobic drugs. As for hydrophilic drugs, biomacromolecules with larger molecular weights exhibited higher encapsulating efficiency than small molecule drugs. In addition, the drug release profiles showed molecular weight-dependent in our study, the macromolecules appeared sustained release and avoid burst release. In summary, due to the efficient loading capacity and ideal release behaviors, loading of hydrophilic macromolecules by EXOs might be the most promising direction for the application and transformation in the field of oral administration. Meanwhile, our study might provide some helpful references for the subsequent studies on EXOs as drug delivery systems.

Conflicts of interest

The authors declare that there is no conflicts of interest.