3.1. Preparation and Characterization of Nanoparticles
Three different drug nanoparticle formulations with the z-average particle size less than 100 nm and PDI < 0.3 were prepared by using the MIVM. The PEGylated polymer, TPGS, was chosen as the primary stabilizer because of its amphiphilic and biodegradable natures. It is a GRAS-listed excipient for pharmaceutical formulations. ITZ, VitD3, and FLU were selected as the cargo because they possess various physicochemical properties (
Table 1). Only FLU was slightly soluble in water (22.0 ± 5.7 μg/mL), while ITZ and VitD3 were practically water insoluble as their experimental solubilities were lower than the LOD of the HPLC (i.e., 0.09 μg/mL for ITZ and 0.05 μg/mL for VitD3). The stream flowrates remained the same for all sample preparation, while the formulation parameters such as organic solvent, initial drug concentration, drug-TPGS ratio, and co-stabilizer, varied in order to generate nanoparticles with adequate particle size (<100 nm) and stability for subsequent analysis. The preparation method of the ITZ and FLU nanoparticles were based on the previous research works [
29,
34].
Although the Z-average particle sizes of these nanoparticles were different, where the ITZ nanoparticle was the largest (93.75 ± 4.33 nm), the VitD3 nanoparticle was the second largest (41.62 ± 3.49 nm), and the FLU nanoparticle was the smallest (28.18 ± 4.22 nm), the PDIs of all the nanoparticles were below 0.3, indicating a narrow size distribution of each formulation (
Figure S1). The ITZ nanoparticle exhibited a nearly null zeta potential, suggesting that most of the ITZ was encapsulated as the zeta potential would tend to be positive if ITZ was present on the surface of the nanoparticle. The zeta potential of the VitD3 nanoparticle was also neutral (
Table 3). However, the FLU nanoparticle exhibited a negative zeta potential, which might be due to the presence of FLU on the surface. The FLU could be free FLU, since the EE of FLU nanoparticles was only 86.06 ± 1.22%, as well as the FLU presented on the surface of nanoparticles. The experimental DLs of all the nano-formulations were consistent to the theoretical values, whereas only ITZ and VitD3 were fully encapsulated by the stabilizers.
To examine the physical stability of all the nanoparticles, the particle size and PDI of the nanoparticles were regularly monitored. In order to ensure all the nanoparticles were stable prior to the release study, various co-stabilizers were employed. CLT was added in the organic stream during the preparation of ITZ nanoparticles and VitD3 nanoparticles because its highly hydrophobic nature could accelerate the molecular rearrangement process of amphiphilic polymer, leading to more hydrophilic block, Polyethylene glycol (PEG), present on the surface [
27]. For the FLU nanoparticles, an aqueous solution containing 0.05%
w/v PVA was employed as the antisolvent to increase the physical stability as the steric hindrance can be enhanced by the hydrophilic layer formed by PVA. In terms of the physical stability, VitD3 nanoparticles (7.00 ± 4.58 h) and FLU nanoparticles (0.67 ± 0.29 h) were limited by the increase of particle size, while that of ITZ nanoparticles (6.00 ± 0.00 h) were restricted by the observation of visible precipitants (
Figure 4 and
Table 3). Not surprisingly, the physical stability of nanoparticles was positively correlated with the lipophilicity of the entrapped drug (R
2 = 0.910).
3.2. Validation of the Separation Methods
To minimize the analytical error, the drug loss during filtration and centrifugal ultrafiltration were determined before the experiments using these techniques. For all the drugs, there was no significant difference between the concentrations of the unprocessed solution and the filtered solutions (
p > 0.05: 0.695 for ITZ, 0.752 for VitD3, 0.165 for FLU) and the absolute variations in assay were less than 0.5% (
Table 4). Thus, the drug loss was considered negligible for subsequent data analysis.
Although centrifugal ultrafiltration has been widely used to separate the free drug from nanoparticles, the impacts of centrifugal speed and duration on the stability of the nanoparticle remain unexplored and these could significantly affect the accuracy of the data obtained such as EE, DL, and even the release kinetics [
20,
35,
36]. Since the FLU nanoparticle was the least stable among all the three formulations, it was selected as the model for the validation of the centrifugal ultrafiltration method. While the same volume of the nanoparticles was loaded into the filter device for all groups, they were centrifuged at different speeds and durations. The intensity of the light scattering in the filtrate was measured using DLS to access the efficiency of the centrifugal ultrafiltration under different processing conditions. The intensity of the light scattering in a solution or a suspension represented its light scattering strength and the light scattered by the ultra-purified water (UPW) was 62 ± 27 cps. Because the scattered light is closely related to the number and the size of the particles in sample, a higher intensity than that of the UPW could serve as an indicator of the presence of the nanoparticles in the filtrate [
37]. In other words, if the intensity of the light scattering in the filtrate was significantly larger than UPW, it would imply an incomplete separation of the nanoparticles and the free drug.
When the centrifugal duration was fixed at 20 min, the centrifugal speed showed minimal impact on either the EE of the nanoparticle (
p = 0.486 > 0.05,
Figure 5A) or the intensity of the light scattering in the filtrate (
p = 0.388 > 0.05,
Figure 5C). On the other hand, when the centrifugal speed was fixed at 4000×
g, neither the EE of the nanoparticle (
p = 0.107 > 0.05,
Figure 6A) nor the intensity of the light in the filtrate (
p = 0.906 > 0.05,
Figure 6C) was significantly altered with the rise of the centrifugal duration. This revealed no substantial damage to the nanoparticle caused by the centrifugal ultrafiltration when the speed and the duration increased. All the intensities of the filtrates obtained under varying speeds and durations were not significantly different from that of the UPW (
p = 0.813 > 0.05), indicating an absence of nanoparticles in the filtrate. However, compared to the unprocessed nanoparticles, the particle sizes after centrifugal ultrafiltration were all larger (
p < 0.0001). A trend was clearly observed that a higher centrifugal speed and a longer centrifugal duration could result in a larger particle size of the resuspended concentrate, which could be attributed to the aggregation of nanoparticles during centrifugation. When the nanoparticles were centrifuged at 4000×
g for 30 min, precipitates were even observed. The volume of the filtrate was also increased with the increase of the centrifugal speed and duration, but the volume would remain the same if the sample was centrifuged at 4000×
g for longer than 20 min. As the concentrate and the filtrate had the same concentration of the free drug, more volume of concentrate implied the presence of more amount of free drug with the nanoparticles. Thus, 15 mL of nanoparticles was centrifuged at 4000×
g for 20 min for EE and DL determinations to ensure complete separation of the free drug and nanoparticles. Under such condition (
Figures S2–S4), the EEs of both the ITZ nanoparticle and the VitD3 nanoparticle were determined as >99.96%. The intensities of the filtrates were not significantly different from that of the UPW (
p = 0.789, 0.189 > 0.05), and the volumes of the filtrates collected from all the three nanoparticles were essentially the same as 14.6 mL. While the particle size of ITZ nanoparticles significantly increased to 136.4 ± 11.7 nm (
p < 0.0001), the VitD3 nanoparticle only became a slightly larger, with a particle size of 45.7 ± 1.5 nm (
p = 0.004 < 0.01).
For the release study, since only a limited volume of sample can be drawn from the release medium, 5 mL of sample will be taken for centrifugal ultrafiltration at 1000×
g for 5 min. Under this condition (
Figures S2–S4), 1.5 mL of the filtrate was collected for all the nanoparticles for HPLC assay of free drug content. The particle size of the FLU nanoparticle increased from 28.18 ± 4.22 nm to 66.0 ± 15.8 nm (
p < 0.0001), and the particle size of the VitD3 nanoparticle also slightly increased from 41.62 ± 3.49 nm to 46.2 ± 2.7 nm (
p = 0.002 < 0.01), while the particles size of the ITZ nanoparticle was not altered significantly (
p = 0.568 > 0.05). However, the encapsulation efficiency of the nanoparticle and the intensity of the light in the filtrate still remained unchanged (
p > 0.05), indicating no nanoparticles were damaged or penetrated through the filter membrane. Therefore, this condition was considered appropriate to separate the nanoparticles and the free drug during the in vitro release study.
3.3. Release Profiles of Nanoparticles
All the release studies were conducted under sink conditions because a non-sink condition could substantially underestimate the drug release from nanoparticles [
38]. Hence, the solubilities of the drugs in their respective release medium were determined prior to the release study. The selection of release medium was based on the USP dissolution database, where 0.1 M HCl solution, 0.1% SDS
w/v solution, and PBS (pH = 7.4) were employed as the release medium for ITZ, VitD3, and FLU, respectively [
31]. ITZ was much less soluble than VitD3 and FLU in the release medium, while FLU was the most soluble (
Table 1). According to their solubilities, different volumes of nanoparticles with equal amount of drug (i.e., 1.25 mg) were loaded into the dialysis tube with 450 mL release medium outside, and a double amount of nanoparticles was added in USP apparatus II (Paddle) with 900 mL release medium. The sink condition was maintained throughout the release studies.
The release profiles of ITZ nanoparticles and the dissolution profile of raw ITZ are plotted in
Figure 7A. The SS methods using syringe filters (i.e., SS + 0.45 µm filter and SS + 0.2 µm filter methods) substantially overestimated the ITZ release owing to the incapability of separation of the free drug and the nanoparticles. As shown in
Figure S1A, there was a small fraction of ITZ nanoparticles larger than 0.2 and 0.45 μm so that they could not pass through the filters, resulting in less than 100% ITZ release was observed in the release study. Although the fractions of ITZ released at 6 h were not significantly different between the SS method using centrifugal ultrafiltration (SS + CU) and DM method (
p = 0.166 > 0.05), the time for commencement of ITZ release from nanoparticles varied. The release of ITZ started after 5 min when using the DM method but 15 min when using the SS + CU method. After 3 h, the release profile of ITZ nanoparticles determined by SS + CU plateaued at around 28%, while more ITZ was continuously released to the receiver compartment in the DM group. Therefore, compared to SS + CU, DM method showed a faster and larger ITZ release. Generally, DM method is considered to underestimate the drug release rate because the observed release rate of the drug in the acceptor compartment can be dictated by its diffusion rate across the dialysis membrane, which may mask the true release rate of the drug from the nanoparticles [
13,
14,
37]. It could also be observed in our study that the concentrations of the free drug in the donor compartment were significantly higher than those in the receiver compartment at 1 h and 3 h (
p < 0.0001) (
Table S1). Even though, DM method appeared to overestimate the release kinetic of ITZ nanoparticles in the present study. This could be due to the leakage of nanoparticles induced by the low compatibility of the dialysis membrane against acidic conditions (
Figure 8A). Hence, the intensity of the light scattering in the nanoparticles in the donor compartment had already been too low (1972 ± 195 cps) to detect the particle size after 1 h, implying only few nanoparticles left within the dialysis membrane (
Table S1). While changing the release medium to PBS (pH = 7.4), the intensity of the light scattering in the nanoparticles in the dialysis tube remained as high as the original sample (12,752 ± 305 cps) and the particle size was increased to 104.8 ± 0.13 nm even after 6 h. In order to compare the repeatability of SS + CU and DM methods, the relative standard deviation (RSD) was calculated for each time point. The RSD of the release profile obtained by the SS + CU method was much lower than that determined by the DM method (
p = 0.019 < 0.05), suggesting a higher repeatability of the SS + CU method.
As with the ITZ nanoparticle, the syringe filters cannot effectively separate the VitD3 nanoparticles, causing a pseudo burst release of VitD3 (
Figure 7B). As no VitD3 nanoparticles were beyond 0.45 μm, but a few of the nanoparticles were larger than 0.2 μm, VitD3 displayed a complete release for the case of using 0.45 μm filters but only around 90% using 0.2 μm filters (
Figure S1B). For the DM method, the initial VitD3 release from the nanoparticles was observed at 5 min. Subsequently, VitD3 was released slowly, attaining around 7.9% of release after 6 h. For the SS + CU method, the nanoparticle started drug release at 15 min. Meanwhile, the release kinetic of VitD3 nanoparticles also plateaued at approximately 1.7%, which was significantly lower than that of the DM group (
p < 0.0001). Hence, both the release rate and the fraction of release determined by the DM method were significantly higher than that of the SS + CU group. Since the centrifugal ultrafiltration technique has been validated to ensure no leakage of drug from nanoparticles and no drug absorbed into the filter membrane, the SS + CU should not misestimate the release kinetic of VitD3 nanoparticles. In other word, the DM method overestimated the release kinetic of the nanoparticles once again, probably because of the extra damage to the nanoparticles. Previous studies found that cellulose could interact with PEG so that the nanoparticles within the dialysis membrane tended to accumulate to the surface of the dialysis membrane gradually during the release study, causing particle aggregation and subsequent destabilization (
Figure 8B) [
39,
40]. As a result, the VitD3 nanoparticles in the donor compartment became unstable after 3 h with a PDI > 0.3, and more VitD3 was released to the receiver compartment (
Table S1). Regarding the RSD, the SS + CU group were again much lower than those of the DM group (
p < 0.0001).
As shown in
Figure 7C, both 0.45 and 0.2 μm filters could not separate free FLU and FLU nanoparticles because of the relatively smaller particle size of FLU nanoparticles (
Figure S3C). A complete release of FLU nanoparticles thus was seen at 5 min. Unlike the ITZ nanoparticle and the VitD3 nanoparticle, FLU nanoparticles released rapidly at the very initial stage regardless of the method used. To compare the release profiles determined by DM and SS + CU methods, the initial release rate (IRR) was calculated by the following equation:
where
is the slope of the initial linear region of the cumulative release curve and
A is the specific surface area of the nanoparticles. As the same FLU nanoparticle formulation was used, the IRR ratio of the SS + CU group to the DM group could be simply obtained by the ratio of their slope, which was 3.3, implying the initial release rate of the SS + CU group was 3.3 times faster than the DM group. Since the FLU nanoparticle became unstable at room temperature within 1 h, it was questionable that the free FLU was gradually released from the nanoparticle over 3 h under a sink condition. In addition, when using the DM method, the average fraction of FLU release observed at 5 min was below 14%, which was the amount of free FLU already existing in the nanoparticles. According to the
Table S1, after 1 h of the release study, all the FLU has already been released out, but the majority of them still stayed within the donor compartment. All of these suggested that the dialysis membrane substantially delayed the translocation of the released FLU to the sampling compartment, leading to an underestimated release kinetic obtained by the DM method (
Figure 8C). On the contrary, the SS + CU group showed a rapid initial release followed by a plateau of 100% after 45 min, which correlated well with the physical stability of the FLU nanoparticle.
In addition to the release profiles, mathematical modeling was applied to study the release kinetics obtained by the DM and SS + CU methods. In this study, two mathematical models including Korsmeyer-Peppas model and Baker-Lonsdale model were selected for comparative analysis, which were commonly used for mathematical modeling of drug release from polymeric nanoparticles [
41,
42]. The former is commonly employed to study the release kinetics of polymeric dosage forms without known mechanisms. The release exponent (n) is used to characterize the release mechanism, where n = 0.43 represents a Fickian diffusion and 0.43 < n < 0.85 indicates a non-Fickian transport for spherical systems [
43]. It should be noted that only data points with less than 60% release can be used for model fitting. The latter model is derived from the Higuchi model for spherical polymeric systems [
44]. As shown in the
Table 5, both models fitted all the data well with high coefficients of determination (R
2 > 0.9). Interestingly, the release exponents of the DM groups were all larger than those of the SS + CU groups, indicating the release kinetics determined by the SS + CU method were more prone to follow the quasi Fickian or Fickian diffusion. This was because the diffusion of drug from the core to the particle surface required penetration through the stabilizers. However, it is worth mentioning that both Korsmeyer-Peppas model and Baker-Lonsdale model were empirical models primarily developed to explain the drug release from conventional controlled release dosage forms [
42]. A more suitable empirical model for release kinetics of nanoparticles should be developed based on the release data collected by SS + CU method accordingly.
It should be noted that the release rates of drug-loaded polymeric nanoparticles were not always faster than the dissolution rates of raw drugs (
Figure 7). Among the three drugs, FLU had the highest dissolution rate and fraction dissolved, while VitD3 possessed faster dissolution rate than ITZ. Such that, the dissolution rate was positively associated with the solubility of the raw drug in the release medium (R
2 = 0.999). On the other hand, the release rates and fractions of drug released followed the descending order as FLU > ITZ > VitD3, which were reversely related to their physical stability. The nanoparticles possessing higher physical stability exhibited a lower release rate (R
2 = 0.995) and fraction of release (R
2 = 0.993). In addition, the fractions of drug released were also negatively correlated to the lipophilicity of the model drugs (R
2 = 0.951). Therefore, rather than improving the solubility of the hydrophobic drugs, stable nanoparticles were more prone to protect the loaded cargos to achieve a sustained release.