Poly(vinylidene fluoride) (PVDF) membrane (MWCO=100 kD) was purchased from AMFOR. Potassium hydroxide (KOH), potassium permanganate (KMnO₄), methanol, dichloromethane, ethanol, acetone, n-hexane, potassium dihydrogen phosphate (KH₂PO₄), dipotassium hydrogen phosphate (K₂HPO₄) and glucose were supplied by local suppliers. Bovine insulin and N-(3-dimethylaminopropyl)-N’-ethylcarbodiimide hydrochloride (EDC) were provided by Yuanye. 2-Aminophenylboronic acid (2-APBA) and bovine serum albumin (BSA) were provided by Bi De Pharm. Fetal bovine serum (FBS) was obtained from Solarbio. Alizarin Red S (ARS) and fluorescein isothiocyanate (FITC) was purchased from Sigma-Aldrich. Azodiisobutyronitrile (AIBN) and cyano-4-(dodecyl thioalkyl thiocarbonyl) thioalkyl valerate (CDTPA) were purchased from Macklin. N-isopropylacrylamide (NIPAM) and N-isopropylmethylacrylamide (NIPMAM) were all obtained from TCI. NIPAM and NIPMAM were purified by recrystallization. Dulbecco’s modified Eagle’s medium/nutrient mixture F-12 (DMEM/F-12), penicillin/streptomycin (P/S) were purchased from Life Technologies. CellTiter 96^® AQueous One Solution Cell Proliferation Assay (MTS) was purchased from Promega. 2-Acrylamidophenylboronic acid (2-AAPBA) was synthesized according to the literature and characterized by ¹H-NMR (Fig. S1 in the electronic supplementary information, ESI).

Before use, PVDF membrane was soaked in anhydrous ethanol for 10 min, cleaned with deionized water and vacuum dried at room temperature to remove surface impurities. Next, the cleaned PVDF membrane was immersed in 4 wt% KOH/alcohol solution at 60 °C for 6 min. Thoroughly clean the membrane with deionized water to terminate the reaction. This step allows the membrane surface to be modified with double bonds. Then, the membrane was immersed in 4 wt% KMnO₄/30 wt% KOH aqueous at 25 °C for 10 min. Clean thoroughly with deionized water to terminate the reaction.^[

CDTPA was modified on the membrane surface through esterification reaction. M-OH with a diameter of 5 cm was added into CDTPA/DCM solution. In order to adjust density of polymer chain, CDTPA concentrations were set as 0.07, 0.14 and 0.56 mmol·L⁻¹. After 10 min, EDC/DCM solution (n_EDC = 10n_CDTPA) was added dropwise to the above solution. The reaction was carried out at 25 °C for 48 h. After the reaction, the membrane was washed with DCM, ethanol and water in turn.^[

Contraction-type PBA-based glucose-sensitive linear polymer was modified on the membrane surface through surface-initiated RAFT polymerization. In 250 mL Schlenk flask, 2-AAPBA, NIPAM and NIPMAM were dissolved in 20 mL of 95 vol% methanol/water. The ratio among NIPAM, NIPMAM, 2-AAPBA was set as 10:5:4. In order to control the chain length, the total monomer concentrations were adjusted to 1.2, 2.4 and 3.6 mmol·L⁻¹. Then, M-CDTPA and the initiator AIBN (0.012, 0.024 and 0.036 mmol·L⁻¹) were added. The reaction was carried out in N₂ at 70 °C for 24 h. The membrane was washed with ethanol and water in turn. After vacuum drying, the membrane modified with glucose-sensitive polymer was obtained.^[

Table 1  Dosage of monomer and CDTPA in glucose-sensitive membrane.

	CDTPA (mmol·L−1)	NIPAM (mmol·L−1)	NIPMAM (mmol·L−1)	2-AAPBA (mmol·L−1)
M-0	/	/	/	/
M-RA0.56-P1.2 a	0.56	12.45	6.20	4.95
M-RA0.56-P2.4	0.56	24.90	12.40	9.90
M-RA0.56-P3.6	0.56	37.35	18.60	14.85
M-RA0.07-P2.4	0.07	24.90	12.40	9.90
M-RA0.14-P2.4	0.14	24.90	12.40	9.90

^a The glucose-sensitive membrane is named as M-RA_x-P_y, where x represents the concentration of CDTPA during the reaction and y represents the total concentration of monomers during the polymerization.

Download:  CSV

The chemical structure of membrane surface was characterized by attenuated total refraction-fourier transform infrared spectrometer (ATR-FTIR, Nicolet iS50, ThermoFisher). The elemental composition of the membrane surface was characterized by X-ray photoelectron spectroscopy (XPS, K-alpha, ThermoFisher). Relative elemental content was analyzed by elemental analyzer (EA, Vario EL Cube, Elementar). The surface morphology was characterized by scanning electron microscopy (SEM, S-4800, Hitachi). The dynamic contact angle of the membrane was measured at room temperature using a contact angle measuring instrument (JYSP-180, Jinshengxin). The surface roughness of the membrane was detected by atomic force microscopy (AFM, Dimension Icon, Bruker) in tapping mode. The porosity of the membrane was characterized by wet-dry weight method. The mass of the membrane was measured by analytical balance (ME204E, Mettler Toledo).

The grafting degree of monomers on the membrane is the increase of the mass of glucose-sensitive membrane in unit area compared to M-CDTPA.^[

$$\mathrm{G}\mathrm{D}=\frac{W_1-W_0}{A}$$

1

where w₀ (μg) and w₁ (μg) are the weight of the M-CDPTA and glucose-sensitive membrane, respectively, A (cm²) is the surface area of the membrane. Weight measurements were conducted for 3 times, and the average value was reported.

M-0 and M-RA_0.56-P_2.4 (1 cm × 1 cm) were immersed in 0.1 mg/mL ARS (50 mmol·L⁻¹ pH 7.4 Tris-HCl) and incubated at 25 °C for 6 h. Then the membrane was washed for 3 times with 50 mmol·L⁻¹ pH 7.4 Tris-HCl. After vacuum drying, the ARS stained membrane was observed by confocal laser scanning microscope (CLSM, FV3000, Olympus).

The cross-flow membrane filtration device was used to test the transmembrane flux at 37 °C and 0.1 MPa. The effective diameter of the membrane is 3 cm. Filtration at 0.1 MPa for 3 h to achieve a stable flux. Then the glucose-sensitivity of the membrane was tested by measuring the glucose solution flux at 37 °C. The glucose solution was prepared in 20 mmol·L⁻¹ pH 7.4 PBS, and the glucose concentrations were 0, 10, 20, 30, 40 and 50 mmol·L⁻¹, respectively. The flux (J) is calculated according to Eq. (2):

$$J=\frac{V}{A\times t}$$

2

where V (L) is the volume of the permeate solution, A (m²) is the area of the membrane, and t (h) is the time. Through three parallel experiments, the average value is recorded as the flux of each membrane.

Insulin permeation was tested with a standard side-by-side diffusion cell at 37 °C (Fig. S2 in ESI). Insulin solution (3 mg·mL⁻¹, 20 mmol·L⁻¹ pH 7.4 PBS with different glucose concentrations) was added in donor cells. Glucose solution (20 mmol·L⁻¹, pH 7.4 PBS) with different glucose concentrations (0, 0.001, 0.002, 0.004 g·mL⁻¹) were added into receptor cell. The glucose concentration in the donor cell is equal to that in the receptor cell. The concentration of insulin in the receptor cell was determined by UV-Vis spectrophotometer (UH4150, Hitachi) at 276 nm.

The diffusion permeability coefficient (P) is calculated according to Eq. (3):

$$P=\frac{M_{\text{t}}\times h}{S\times C_{\text{d}}\times t}\times 100\text{\%}$$

3

where M_t (mg) is the mass of drug permeation at time t (s), h (cm) is the thickness of the membrane, S (cm²) is the effective permeation area, C_d (mg/mL) is the insulin concentration in donor cells, and t (s) is the permeation time.

In order to study whether the inorganic salts and other metabolites in the blood have an effect on the glucose-sensitivity of the membrane, insulin permeation test was also carried out in simulated human body fluids (SBF) and fetal bovine serum (FBS). The formula for preparing SBF is as follows: NaCl (8.035 g), NaHCO₃ (0.355 g), K₂HPO₄·3H₂O (0.225 g), MgCl₂·6H₂O (0.311 g), 1.0 mol·L⁻¹ HCl (39 mL), CaCl₂ (0.292 g), Na₂SO₄ (0.072 g), Tris (6.118 g), 1.0 mol·L⁻¹ HCl (1−5 mL) were sequentially added into 700 mL H₂O at room temperature. The pH of solution was adjusted to 7.4, with 1.0 mol·L⁻¹ HCl or 1.0 mol·L⁻¹ NaOH. Then the insulin release test of the membrane in SBF was tested.

The insulin release of the membrane in FBS was tested according to this method. The PBS was replaced with FBS. Insulin was replaced by fluorescein isothiocyanate labeled insulin (FITC-insulin).^[

The membrane was pre-wetted with 20 mmol·L⁻¹ pH 7.4 PBS. Then, it was immersed in 40 mL of 0.5 mg·mL⁻¹ BSA solution (20 mmol·L⁻¹ pH 7.4 PBS) for 12 h to achieve adsorption equilibrium. The membrane was washed with deionized water to remove BSA from the surface. Then, it was subjected to ultrasonic treatment in 40 mL of 20 mmol·L⁻¹ pH 7.4 PBS for 1 h. The protein concentration of the eluent obtained after ultrasound was determined by UV-Vis. The same membrane was tested for 3 times, and the results were averaged. The adsorption capacity of BSA (Q, μg·cm⁻²) was determined by Eq. (4):

$$Q=\frac{\mathrm{\Delta }C\times V}{S}\times 100\text{\%}$$

4

where ΔC is the difference between the initial concentration of BSA and the concentration after adsorption equilibrium, V (mL) is the volume of BSA solution, and S (cm²) is the area of the membrane.^[

Firstly, the membrane was cut into 1 cm × 1 cm samples, and then immersed in ethanol for 30 min for disinfection. At 37 °C, L929 mouse fibroblast-like cells were cultured in DMEM/F-12 medium containing 1% antibiotics and 10% FBS, and the external environment was humid atmiosphere containing 5% CO₂. L929 cells were seeded in 24-well plates at a density of 5×10⁴ cells per well. Cells cultured without the membrane were used as controls. Cell viability was evaluated after 24, 72 and 120 h of culture. MTS assay was used to evaluate cell viability. Absorbance at 490 nm was recorded with a VICTOR^® NIVO^TM mutlimode plate reader (PerkinElmer, USA).

In order to test the stability of glucose-sensitive membrane. The glucose-sensitive membranes were immersed in 20 mmol·L⁻¹ pH 7.4 PBS at 37 °C and shaken for 7 h. Glucose-sensitive flux was measured before and after the treatment to evaluate the stability of the membrane. In addition, the glucose-sensitive membrane immersed in PBS for 7 days was subjected to a flux test in a glucose concentration environment of 0 and 50 mmol/L alternating cycles to further test the stability of the membrane.^[

The rejection rate of the membrane was tested by cross-flow membrane filtration device. The solutes were 500 mg·L⁻¹ BSA (6.7×10⁴ Da), ovalbumin (4.3×10⁴ Da), and insulin (5.8×10³ Da), respectively. Before the test, the protein solution concentration-absorbance standard curve was drawn. The membrane was first compacted with pure water for 30 minutes to eliminate any effect of compaction on flux decline. Then, the protein rejection rate of M-RA_0.56-P_2.4 in 0, 10, 20, 30, 40 and 50 mmol·L⁻¹ glucose (in 100 mmol·L⁻¹ pH 7.4 PBS) was tested at 37 °C. The rejection rate is calculated by Eq. (5):^[

$$R=\frac{C_0-C_1}{C_0}\times 100\text{\%}$$

5

where C₀ is the initial concentration of each protein and C₁ is the concentration of each protein after passing through the membrane.

Glucose-sensitive polymer was grafted on the membrane surface through surface-initiated RAFT polymerization. Briefly, PVDF membrane was treated with KOH and KMnO₄ to introduce hydroxyl groups on the membrane surface. Then, the RAFT agent (CDTPA) was modified on the membrane surface through esterification reaction. Finally, glucose-sensitive polymer chain was grafted on the membrane surface by RAFT polymerization (Fig. S3 in ESI).

ATR-FTIR was used to characterize the chemical structure of the membrane surface. As shown in Fig. 1(a), ATR-FTIR spectrum of pristine PVDF membrane (M-0) shows three strong characteristic bands at 1400, 1167 and 880 cm⁻¹, which can be attributed to ―CH₂―, ―CF₂― stretching and C―C bond skeleton vibration, respectively. After alkali treatment, a strong absorption band appears at 3400 cm⁻¹. This is the characteristic band of ―OH, which proves that ―OH has been successfully introduced to the membrane surface. After esterification, the band strength of ―OH characteristic band decreases significantly, proving that CDTPA was successfully modified. The characteristic absorption band of amide bond (1550 cm⁻¹) appears after surface-initiated RAFT polymerization, demonstrating that glucose-sensitive polymer was successfully modified on the membrane surface.^[

Fig 1  (a) ATR-FTIR spectra of M-0, M-OH, M-CDTPA and M-RA_0.56-P_2.4; (b) XPS spectra of M-0, M-OH, M-CDTPA and M-RA_0.56-P_2.4.

Download:  Full-size image | High-res image | Low-res image

The chemical structure of the membrane surface was further confirmed by XPS (Fig. 1b, Table S1 in ESI). The XPS of M-0 shows emission peaks corresponding to C 1s, N 1s, O 1s and F 1s at 285.27, 400.09, 532.07 and 688.23 eV, respectively. M-0 contains N and O, which may be due to the addition of additives in the process of membrane preparation. Compared with M-0, the content of F element of M-OH decreased from 37.38% to 8.13%. This is due to the removal of HF on the membrane surface after alkali treatment. Due to the introduction of hydroxyl group on the membrane surface, the content of oxygen element on the membrane surface increased from 3.29% to 15.75%. After CDTPA modification, S element (168.2 and 232 eV corresponding to S 2p and S 2s) appeared on the membrane surface, which proved that CDTPA was successfully modified on the membrane surface. After polymerization, the content of N element (400.04 eV) increased from 3.79% to 9.21%, while the content of S element decreased from 2.57% to 1.53%. This indicated that the polymer was successfully grafted onto the surface of the membrane. The XPS results were consistent with the ATR-FTIR results.

In order to further verify glucose-sensitive polymer was successfully grafted on the membrane surface, the membrane was stained with ARS and observed with CLSM. PBA can combine with diol on ARS to form PBA-ARS complex, which can improve the fluorescence intensity of ARS (Fig. S4 in ESI).^[

Polymer chain length and chain density have significant effect on glucose-sensitivity of the membrane.^[

In order to adjust the polymer chain length, we fixed the concentration of CDTPA as 0.56 mmol·L⁻¹, and changed the total concentration of monomer during polymerization. The total concentration of monomer was set as 1.2, 2.4 and 3.6 mmol·L⁻¹. The resulting membrane was named as M-RA_0.56-P_1.2, M-RA_0.56-P_2.4 M-RA_0.56-P_3.6. Since the polymer was fabricated from CDTPA active sites, the density of the polymer chain on the membrane surface was equal to the density of CDTPA on the membrane surface before polymerization. Therefore, M-RA_0.56-P_1.2, M-RA_0.56-P_2.4, M-RA_0.56-P_3.6 have similar chain densities. Under the same density of CDTPA and the same polymerization conditions, with the increase of monomer concentration, the length of polymer chain increases. As a result, the grafting degree of the membrane increases (Fig. 2a). ATR-FTIR and XPS results confirmed that the amount of grafted polymer increased with the increase of monomer concentration (Fig. S6 in ESI). By adjusting the total concentration of monomer, we obtained membranes with similar density and different chain lengths.

Fig 2  (a) Grafting degree (GD) and (b) glucose-sensitive flux of M-0, M-RA_0.56-P_1.2, M-RA_0.56-P_2.4, M-RA_0.56-P_3.6.

Download:  Full-size image | High-res image | Low-res image

The effect of polymer chain lengths on the glucose-sensitivity (glucose sensitivity coefficient R, the ratio of membrane flux to 0 mmol/L membrane flux in a 50 mmol·L⁻¹ glucose environment) of the membrane was investigated (Fig. 2).^[

Next, the graft density of the polymer chain was adjusted by changing the concentration of CDTPA, which was set as 0.07, 0.14 and 0.56 mmol·L⁻¹, respectively.^[

Fig 3  (a) Grafting degree (GD) and (b) glucose-sensitive flux of M-0, M-RA_0.07-P_2.4, M- RA_0.14-P_2.4, M-RA_0.56-P_2.4.

Download:  Full-size image | High-res image | Low-res image

The effect of chain density on membrane surface on glucose-sensitivity was explored. As shown in Fig. 3(b), when the graft density of the polymer chain increases, the flux of the membrane decreases, while glucose-sensitivity of the membrane increases. As the density of the polymer chain increases, the number of polymer chains acting as chemical valves increases (Figs. S8B and S8D in ESI). By adjusting CDTPA and total monomer concentration, M-RA_0.56-P_2.4 has the optimal glucose-sensitivity. In order to more effectively prove the glucose sensitivity of the membrane, the rejection rate of the membrane was tested with BSA, ovalbumin and insulin. As shown in Fig. S9 (in ESI), at each glucose concentration, the rejection rate decreased with the decrease of molecular weight. For each protein, the rejection rate decreased with the increase of glucose concentration. These results were consisting with the flux results. The results confirmed that insulin can permeate through membrane under pressure.

In order to observe the effect of surface grafting on membrane microstructure, the surface of the membrane was observed by SEM. Membrane pores can be clearly observed on the membrane surface of M-0 (Fig. 4a), M-OH (Fig. 4b), and M-CDTPA (Fig. 4c). Furthermore, the size of pores on the membrane surface does not change significantly. After polymerization, membrane surface became denser, and the size of pores decreases significantly (Figs. 4d−4i). With the increase of chain density, the size of pores on the membrane surface becomes smaller and the membrane surface becomes denser (Figs. 4d−4f). The same is true as the polymer chain length increases (Figs. 4g−4i). This is due to the fact that the grafted polymer covers the pores on the membrane surface. The roughness of the membrane was characterized by AFM. As the polymer chain density or chain length increases, the roughness of the membrane increases (Fig. S10 in ESI). The results were consistent with the SEM results.

Fig 4  SEM images the surface of (a) M-0, (b) M-OH, (c) M-CDTPA, (d) M-CDTPA_0.07-P_2.4, (e) M-CDTPA_0.14-P_2.4, (f) M-CDTPA_0.56-P_2.4, (g) M-CDTPA_0.56-P_1.2, (h) M-CDTPA_0.56-P_2.4, (i) M-CDTPA_0.56-P_3.6. Scale bar is 0.5 μm.

Download:  Full-size image | High-res image | Low-res image

The hydrophilicity of the membrane was studied by dynamic contact angle (Fig. S11 in ESI). Compared with M-0, the hydrophilicity of M-OH was improved. This was due to the formation of ―OH on the membrane surface. After modification of CDTPA, the hydrophilicity of membrane was deteriorated, due to the hydrophobicity of dodecyl in CDTPA. After polymerization, the hydrophilicity of the membrane surface was improved due to the hydrophilicity of the glucose-sensitive polymer (Fig. S11A in ESI). With the increase of chain length or chain density, the initial contact angle of the membrane decreases gradually. The contact angle of the membrane gradually decreases over time, until the equilibrium is reached (Figs. S11B and S11C in ESI). With the increase of polymer chain length or chain density, the hydrophilicity of the membrane surface increases gradually.^[

Standard side-by-side diffusion cell was used to study the permeability of insulin on the membrane. The glucose concentration was selected to simulate the blood glucose level in the human body, 0.001 g·mL⁻¹ (normal pre-meal blood glucose in non-diabetic patients), 0.002 g·mL⁻¹ (diabetic patients) and 0.004 g·mL⁻¹ (maximum detectable value of a blood glucose test device and test strip).^[

Fig 5  Cummuiative insulin release through (a) M-0 and (d) M-RA_0.56-P_2.4 at different glucose concentrations; Diffusion permeability coefficient (P) of (b) M-0 and (e) M-RA_0.56-P_2.4 at different glucose concentrations; The ratio of insulin diffusion permeability coefficient of (c) M-0 and (f) M-RA_0.56-P_2.4.

Download:  Full-size image | High-res image | Low-res image

Alternating release ability is one of the most important conditions for self-regulating insulin delivery. By alternating the glucose concentration between 0.004 and 0.001 g·mL⁻¹, changes in the amount of insulin released can be observed. The amount of permeated insulin (Fig. 6a) and the permeability coefficient of insulin (Fig. 6b) increases with increasing glucose from 0.001 g·mL⁻¹ to 0.004 g·mL⁻¹ and decreases when glucose is reduced to 0.001 g·mL⁻¹. Similar results can be observed in four consecutive cycles and P₄₀₀/P₁₀₀ hardly changed (Fig. 6c). These results demonstrated the reproducibility of the membrane.

Fig 6  (a) Commutative insulin release and (b) P of M-RA_0.56-P_2.4 at different glucose concentrations in four alternating cycles; (c) The diffusion permeability coefficient ratio (P₄₀₀/P₁₀₀) in four alternating cycles.

Download:  Full-size image | High-res image | Low-res image

In order to further test the release of insulin in the physiological environment, FBS and SBF were used to simulate the physiological environment. In FBS or SBF, the insulin release and P value of M-RA_0.56-P_2.4 still increased with the increase of glucose concentration, as shown in Figs. S13(a), S13(b), S13(d) and S13(e) (in ESI). The glucose sensitivity coefficients (P₄₀₀/P₁₀₀, P₂₀₀/P₁₀₀) in FBS are 1.43 and 1.25, respectively. The glucose sensitivity coefficients (P₄₀₀/P₁₀₀, P₂₀₀/P₁₀₀) in SBF are 1.39 and 1.21, respectively. The above results are similar to those in PBS. These results indicate that protein and some metabolite in human body fluids have no significant effect on the glucose-sensitivity of the membrane. The M-RA_0.56-P_2.4 prepared by this method has good self-regulated glucose-sensitive insulin release at physiologically relevant blood glucose concentrations.

BSA was used for static adsorption experiments to characterize the anti-fouling performance of the glucose-sensitive membrane (Fig. S14 in ESI).^[

To observe the membrane stability, the glucose-sensitive flux of M-RA_0.56-P_2.4 was measured before and after rinsing. The glucose-sensitive flux of the membrane did not change significantly before and after rinsing (Fig. 7a). Furthermore, a reversible glucose-sensitive flux test was performed (Fig. 7b). These results further confirmed that the membrane still had good glucose-sensitivity after rinsing.

Fig 7  (a) Glucose-sensitive flux of M-RA_0.56-P_2.4 before and after rinsing; (b) Reversible glucose-sensitive flux M-RA_0.56-P_2.4 after rinsing.

Download:  Full-size image | High-res image | Low-res image