Microfluidic Synthesis of Rigid Nanovesicles for Hydrophilic Reagents Delivery

We present a hollow-structured rigid nanovesicle (RNV) fabricated by a multi-stage microfluidic chip in one step, to effectively entrap various hydrophilic reagents inside, without complicated synthesis, extensive use of emulsifiers and stabilizers, and laborious purification procedures. The RNV contains a hollow water core, a rigid poly (lactic-co-glycolic acid) (PLGA) shell, and an outermost lipid layer. The formation mechanism of the RNV is investigated by dissipative particle dynamics (DPD) simulations. The entrapment efficiency of hydrophilic reagents such as calcein, rhodamine B and siRNA inside the hollow water core of RNV is ≈90 %. In comparison with the combination of free Dox and siRNA, RNV that co-encapsulate siRNA and doxorubicin (Dox) reveals a significantly enhanced anti-tumor effect for a multi-drug resistant tumor model.

S-3 first stage, or FR below ~15 at the second stage could result in the inefficient mixing inside microchannels, and the precipitated PLGA particles tended to be large in size.
We chose FR of 10 at the first stage (side inlets: 1 mL/hr each, middle inlet: 0.2 mL/hr), and FR of ~15 at the second stage (side inlets: 15 mL/hr each) considering the two following factors. (1) The microfluidic chip that we used was made by bonding replica PDMS with the glass substrate. If we further increased the flow rate, the performance of chip might become unstable over a long working period due to the high pressure induced by fluids. (2) If we increased the FR at the first or the second stage, the final concentration of siRNA inside RNV would be decreased, making it less efficient for the following in vitro and in vivo experiments. The production rate of RNV is ~ 114 µg/min (6.832 mg/hr) by the microfluidic chip. For a RNV of a = 140 nm, and ρ =0.5 g/cm 3 , the calculated mass is ~ 1×10 -15 g, and the number of generated particles is ~ 1.14×10 11 per minute. The amount of RNV generated per hour is enough for 20 doses for one mouse.

Characterization of RNV
The RNV generated with microfluidic chip was characterized with dynamic light scattering(DLS, Zetasizer 3000HS, Malvern Instruments Ltd.) and transmission electron microscopy(TEM, FEI Tecnai T20). The hydrodynamic diameter and zeta potential of particles were measured by DLS at a scattering angle of 173°. The structure of RNV was observed with TEM (acceleration voltage 200 kV) by dropping RNV suspension onto a carbon film-coated grid and then air-dried at room S-4 temperature before imaging.

Characterization on encapsulation of hydrophilic argents by RNV
The fluorescence spectrum of RNV encapsulating calcein and rhodamine B was obtained with spectrofluorophotometer (RF-5301PC, SHIMADZU) with excitation at 455 nm for calcein and 556 nm for rhodamine B. The spectrum of free dyes was recorded at the same wavelength and compared with that of RNV. The results indicated that the dyes were successfully entrapped into the RNV, because the fluorescence spectrum of RNV was the same as that of free dyes ( Figures 1B, 1C, S2, and S3).

Characterization on encapsulation efficiency
For determining encapsulation efficiency, the RNV entrapping hydrophilic molecules (calcein and rhodamine B) was filtered with Amicon Ultra-0.5ultrafilter (MWCO=30KDa, Millipore, USA). The concentration of free calcein or rhodamine B in the diffusate was measured by detecting the fluorescence intensity at 518 nm (excitation wavelength 488 nm for calcein) or 590 nm (excitation wavelength 548 nm for rhodamine B) with infinite M200 microplate reader. The encapsulation efficiency of dyes equals to (Qtotal-Qfree)/Qtotal, in which Qtotal is the total amount of dyes and Qfree is the amount of free dyes in diffusate. The amount of siRNA encapsulated in the RNV and free siRNA were determined by agarose gel electrophoresis assay. The agarose gel after electrophoresis was stained with ethidium bromide and imaged with S-5 Image Quant 3000. The image was analyzed with ImageJ 2x (NIH) to quantify the fluorescence of siRNA bands.

Lyophilization of RNV
For long-term storage, the RNV was suspended in aqueous solution with 1% mannitol and 10% trehalose and lyophilized in vacuum at -60 o C for 24 h with a FD-1A-50 vacuum freeze dryer (Boyikang, Beijing) to remove water and DMF.
There is almost no change of RNV size and polydispersity index (PDI) after lyophilization (Table S2). TEM images of RNV[Dox/siMDR1] before and after lyophilization also verify that lyophilization will not affect the complex structure of RNV ( Figure S4).

In vitro release of Dox
In vitro release of Dox from the RNV[Dox/siMDR] was performed at pH 7.4 and Under the condition of pH 7.4, the release rate of Dox encapsulated by PLGA shell was relatively slow and the total amount of released Dox was below 40 % even after 72 hr ( Figure S5). In comparison, the release rate of encapsulated Dox was faster at pH 4.5. After 48 hr, more than 60 % of Dox was released ( Figure S5). The intracellular pH of tumor cells, especially in lysosomes, is generally below 5, so we speculate that 72 hr is sufficient for the degradation of PLGA. However, we found that it is hard to measure the release curve of siRNA inside the RNV because the siRNA tends to quickly degrade after releasing to the environment. Based on the in vitro gene silencing experiment ( Figure 3C) and the degradation of PLGA, we believed that the release of siRNA occurred within 72 hr.

Dissipative particle dynamics (DPD) simulations
Because the intermediate steps of the formation of the particles and the synthesis process of RNV within the microfluidic channel are difficult to capture and characterize using experimental approaches, we carryout dissipative particle dynamics (DPD) simulations. [2] It is shown that DPD is a very useful method to study biological systems, especially for biomembrane systems. [3] In the DPD simulation, the force on bead i due to bead j is given as a sum of 3 terms: where is a conservative force, is a dissipative force, is a random force, is the distance between beads i and j, and is the cutoff distance. (If exceeds , there would be no interaction between i and j.). The conservative force acts to give beads a chemical identity, while the dissipative and random forces together form a thermostat that keeps the mean temperature of the system constant.
The conservative force between beads i and j is soft repulsive and determined by: where is the maximum repulsive strength, ̂= / is the unit vector.The interaction parameters are shown in Table S3.
We adopt the lipid model developed by Groot and Rabone. [4] In this model, the where spring constant = 50 , and equilibrium bond length = 0.7 . We also apply a harmonic constraint on the adjacent three beads, and details can be seen in the paper by Li et al. [5] Comparing to a typical membrane thickness of 4 nm, the basic length unit, , in the simulation is about 0.8 nm. By mapping the diffusion coefficient around for 5 μm 2 /s for lipids, [6] the time unit = [ 2 /( )] 1 2 ⁄ , is about 24.32 ps.
We use 200 connected beads for PLGA chains. The neighboring beads in PLGA are also connected together by a harmonic spring with spring constant = 200 , and equilibrium bond length = 1.0 .

S-8
Step1: generation of water droplets in PLGA Firstly, we study the process to mix water, lipids, PLGA. The simulation box is a cuboid of size 20 20  110 , with periodic boundary condition applied along x, y and z directions. The system consists of 400 lipid molecules, 613 PLGA chains and 4200 water beads, as shown in Table S4. There are totally 132,000 beads in the system with particle density about 3.0. We put a relaxed water bead into a cylindrical column with a diameter of 4 in the center of the box and other beads outside the water column. At the beginning of the simulation, we fix the water and let other beads relax.
(At this step, to ensure lipids randomly distributed in PLGA, we have set between water and other beads as 100, and among other beads except for water as 25.) Then we release the water and perform DPD simulations. We can see water form droplets quickly under the interfacial tension between water and PLGA ( Figure   2). Afterwards, lipids will assemble at the surface of the water droplets, and form reverse micelles.

Step2:Generation of RNV
Then we study the process when we add more water to the system. We construct a larger system with box size 50  50  110 , and put the as-fabricated system in the first step to the center of the box. We add water bead with beads density of 3.0 to the system. At this moment, the system consists of 825,000 beads in total. Periodic boundary condition is also applied along x, y and z directions. were carried out for 3 times and the results were shown as mean ± SD.

Cell apoptosis
Cell apoptosis was quantified by Annexin V-FITC/PI assay. MCF-7/ADR cells were seeded in 6-well plates and treated with different nanoparticles or free drug in complete medium. The final concentration of siRNA is 100nM while that for Dox is 5 μg/mL. After 72hr incubation, all cells were harvested and stained with Annexin V-FITC and PI following the manufacturer's instructions. Fluorescence of cells was measured using a FACScan flow cytometer (Beckman Quanta SC, US).

In vivo tumor growth inhibition study
Female Balb/c mice (6-8 weeks old), weighing 18-22 g were purchased from Vital River Laboratory Animal Center (Beijing, China). 5×10 6 MCF-7/ADR cells were inoculated subcutaneously(s.c.) in the right flank of the Balb/c nude mice [9] . When tumor size reached 120-150 mm 3 in volume, animals were sacrificed and tumors were aseptically dissected and minced with scissors into 15mm 3 pieces. Then, tumor tissues were transplanted s.c. into the armpit of the mice. When the tumor volume was above 70 mm 3 , mice were randomly divided into 5 groups (5 animals per group): (1) PBS, vein injection. The day before the first dose was specified as day 0. Tumor size was measured daily using a vernier caliper across its longest (L) and shortest (S) diameters and the volume was calculated according to the formula of V = 0.5LS 2 . Body weight of the animals was measured at day 0 and day 12. Two days after the last injection, the animals were sacrificed, and the tumor tissues were weighed.
We have not included the control of RNV[siMDR1] (RNV has the active siRNA against MDR1 but lacks Dox) because the previous investigation showed that nanoparticles encapsulating siMDR1 had no anticancer effects. [10] siRNA against MDR1 is not expected to have a therapeutic effect in the absence of a chemotherapeutic to take advantage of the change in protein translation with siRNA knockdown.
There might be two main reasons for the fact that RNV[Dox/siMDR1] did not eradicate the tumor, but rather just prohibited its growth. (1) The dead tumor cells still occupied some space in the tumor site, making the tumor have the similar size as before treatment. [11] (2) RNV[Dox/siMDR1] might not penetrate or spread through the whole tumor tissue so there was a possibility that some parts of tumors were still growing. This matches the result of some previous work. We did not monitor the animals longer because we observed a significant difference between the RNV[Dox/siMDR1] and other groups at day 10, so we terminated the experiments only two days after the last dose mainly considering the animal welfare in order to reduce the time of animals suffering from cancer. [12] S-15

Statistical analysis
For statistical analysis between two groups, Student's t-test for independent means was applied. The differences between any two groups out of several groups were analyzed by one-way analysis of variance (ANOVA) followed by Tukey multiple comparisons. Statistical analysis was performed with the SPSS software (Version 16.0, SPSS Inc, Chicago). A value of P < 0.05 was considered as statistically significant. siMDR1(sense strand) [13] GAA ACC AAC UGU CAG UGU AdTdT siNC(sense strand) UUCUCCGAACGU GUCACGUdTdT Forward primer β-Actin [13] ACC AAC TGG GAC GAC ATG GA Reverse primer β-Actin [13] CTC CTT AAT GTC ACG CAC GCA CGA Forward primer MDR1 [14] AGG AAG CCA ATG CCT ATG ACT TTA Reverse primer MDR1 [14] CAA CTG GGC CCC TCT CTC TC