Injectable sustained‐release poly(lactic‐co‐glycolic acid) (PLGA) microspheres of exenatide prepared by supercritical fluid extraction of emulsion process based on a design of experiment approach

Abstract This study aimed to develop an improved sustained‐release (SR) PLGA microsphere of exenatide using supercritical fluid extraction of emulsions (SFEE). As a translational research, we investigated the effect of various process parameters on the fabrication of exenatide‐loaded PLGA microspheres by SFEE (ELPM_SFEE) using the Box–Behnken design (BBD), a design of experiment approach. Further, ELPM obtained under optimized conditions and satisfying all the response criteria were compared with PLGA microspheres prepared using the conventional solvent evaporation (ELPM_SE) method through various solid‐state characterizations and in vitro and in vivo evaluations. The four process parameters selected as independent variables were pressure (X 1), temperature (X 2), stirring rate (X 3), and flow ratio (X 4). The effects of these independent variables on five responses, namely the particle size, its distribution (SPAN value), encapsulation efficiency (EE), initial drug burst release (IBR), and residual organic solvent, were evaluated using BBD. Based on the experimental results, a desirable range of combinations of various variables in the SFEE process was determined by graphical optimization. Solid‐state characterization and in vitro evaluation revealed that ELPM_SFEE improved properties, including a smaller particle size and SPAN value, higher EE, lower IBR, and lower residual solvent. Furthermore, the pharmacokinetic and pharmacodynamic study results indicated better in vivo efficacy with desirable SR properties, including a reduction in blood glucose levels, weight gain, and food intake, for ELPM_SFEE than those generated using SE. Therefore, the potential drawback of conventional technologies such as the SE for the preparation of injectable SR PLGA microspheres could be improved by optimizing the SFEE process.

. From this fact, it was suggested that the improper release of exenatide from Byetta ® LAR in vivo may explain the necessity for a much higher dose. 60 Another study reported that the drug burst release was high (approximately 45%) when they prepared microspheres following the commercialized ELPM formulation. 59 These problems can increase the risk of adverse effects due to overdose of exenatide. 54 It was also reported that in vivo sustained PK profile of Byetta ® LAR showed poor SR characteristics with low release between days 4 and 17 in animal PK study. 55 In addition, the commercial coavervation method for the fabrication of PLGA microsphere has several problems such as rather complicated process and long manufacturing time with multiple process steps, the use of additional oil-based dispersion solvents and toxic solvents such as n-hexane to remove other solvent. Various pharmaceutical approaches have been tried to overcome these problems. In particular, several research groups have been attempted to reduce burst release and the incomplete release of exenatide from commercial SR microsphere products. 54,55,57,59,[61][62][63][64][65][66][67][68][69][70][71][72][73][74][75][76] Nevertheless, limitations in the microsphere manufacturing process, physical quality characteristics and drug release property have not been fully overcome.
Fundamental ways to overcome this limitation are to simultaneously try approaches to improve the pharmaceutical properties of SR injection formulations as well as the improvement or development of manufacturing processes for its commercialization. These attempts are not only limited to basic research on a laboratory scale but can drive and provide improved innovative pharmaceutical technology that can impact clinical practice and/or commercial SR injection formulations.
Thus, we also investigated the effect of various process parameters on the fabrication of exenatide-loaded PLGA microspheres by SFEE (ELPM_SFEE) using the Box-Behnken design (BBD), a design of experiment (DOE), as a translational research approach to reduce the gap between fundamental understanding of the SFEE process and its application to scale-up manufacturing of PLGA microsphere for commercialization. To the best of our knowledge, there have been no attempts to investigate the SFEE process for the production of SR microsphere containing exenatide based on the DOE approach. BBD was used as a response surface methodology (RSM) to mathematically and statistically determine the patterns and relationships between independent variables, including the SFEE process parameters (pressure, temperature, and stirring rate) and important responses related to the fabrication of PLGA microspheres, such as EE, PSD, IBR, and residual organic solvent.
In addition, to evaluate critical pharmaceutical quality properties and clinical relevance as a final drug product, various in vitro characterizations, such as scanning electron microscopy (SEM), particle size analysis, drug release, Brunauer-Emmett-Teller (BET) specific surface area measurement, water vapor sorption analysis, Fourier-transform infrared (FT-IR) spectroscopy, circular dichroism (CD) spectroscopy, and evaluation of EE, were also performed in comparison with ELPM prepared by the SE process, which is a common conventional commercial technique for preparing SR microsphere formulations. Further, the feasibility of clinical use for the PLGA microsphere formulation prepared using the optimized SFEE process was evaluated. Plasma exenatide concentrations as a pharmacokinetic (PK) property, as well as nonfasting blood glucose, and changes in food intake and body weight as pharmacodynamic (PD) and therapeutical efficacy properties were evaluated in streptozotocin (STZ)induced diabetic mice following treatment with this novel formulation.
The results of these evaluations were compared with those of PLGA microspheres prepared using the conventional SE method.

| Reverse phase high performance liquid chromatography for exenatide quantification
Reverse phase high performance liquid chromatography (RP-HPLC) was performed using an Agilent 1290 Infinity HPLC system (Germany) to quantitatively analyze the exenatide content, as previously described, with slight modifications. 55,77,78 The HPLC system comprised a high-pressure pump (Model 1260 Quat Pump VL) and an auto sampler (Model 1260 ALS) that was maintained at a constant temperature of 5 C. Exenatide was separated using an octadecylsilane column (Kintex C18, 2.6 μm, 100 Å, 4.6 Â 100 mm 2 , Phenomenex Inc., USA) at a temperature of 30 C. Exenatide was detected at 210 nm using a UV detector (Model 1260 VWD DL). The mobile phase comprised 40% ACN with 0.1% TFA/60% water with 0.1% TFA, eluted at a flow rate of 1.5 ml/min, and the injection volume was set as 10 μl.

| Preparation of W/O/W double emulsion
The formulation used in this study was determined based on the results of our previously reported studies, 54,74,77,78  and gently mixed to obtain system equilibrium.

| SFEE process
PLGA microspheres were prepared using the SFEE process under various process conditions to assess the effect of process variables on the properties of PLGA microspheres. A schematic representation of the SFEE apparatus used for microsphere preparation is shown in Figure 1. 79 First, liquid CO 2 from the storage tank was delivered into the preheater tank; heated and compressed CO 2 was then delivered to the cylindrical stainless-steel extraction vessel (approximately 70 ml) through the spray nozzle using an ISCO syringe pump (Model 260D, USA) until the desired pressure was reached. Once the pressure and temperature were equilibrated under the desired conditions, 15 ml of the W/O/W emulsion was injected into the high-pressure extraction vessel using an HPLC liquid pump (Model 307; Gilson Inc., USA). Thereafter, SC-CO 2 continued to flow into the vessel at a constant flow rate to extract DCM from the double emulsion using a backpressure regulator (Tescom, model 26-1723-24-194, USA) at a constant pressure and temperature. The emulsion was stirred using a magnetic stirrer during extraction for effective diffusion and to prevent coalescence in the emulsion. Each SFEE process was performed for 5 h. After complete the extraction, the vessel was slowly depressurized to atmospheric pressure. The collected PLGA microsphere suspensions were then washed using a sterilized injection solution (0.9% NaCl and 5% dextrose) by centrifugation at 600Â g (Eppendorf centrifuge 5019R, Germany) followed by supernatant removal. PLGA microsphere suspension was washed thrice using sterile deionized water for performing solid-state characterization which will be described in Section 2.5. Finally, the washed wet PLGA microspheres were frozen in a deep freezer and then freeze-dried (FD8508, Ilshin Bio-Base Co. Ltd., Gyeonggi-do, Korea) at a vacuum of 50 mTorr.

| Experimental design and optimization of the SFEE process for manufacturing microspheres using BBD
The effect of critical process parameters in SFEE, alone or in combination, as independent variables on five responses, including the particle size and its distribution (in terms of SPAN value), EE, IBR, and residual organic solvent, were elucidated using BBD using RSM. These responses were selected as critical quality attributes for SR PLGA microspheres based on previously reported literature. [1][2][3][4][5][6][7] Further, based on the results from our preliminary study, the CQA, which can be affected by the SFEE process variables, was also considered for the proper response selection.
For BBD, the experimental methodology and generation of mathematical models were performed using Design Expert software (version 7.0, Stat-Ease Inc., Minneapolis, MN, USA). A BBD with four factors, three levels, and a total of 27 experimental runs, including three center points per block, was selected for RSM. The four process parameters used as independent variables were pressure (X 1 ), F I G U R E 1 Schematic representation of the supercritical fluid extraction of emulsions (SFEE) apparatus. (Adopted with permission by Creative Common CC BY license). 64 temperature (X 2 ), stirring rate (X 3 ), and flow ratio (X 4 ). These chosen factors are well known as critical process parameters for SFE processes in various fields. 17,[80][81][82][83][84][85][86][87] The levels of these four process parameters were determined based on a preliminary study conducted with a wider range of the four independent variables (X 1 : 30-65 C, X 2 : 70-160 bar, X 3 : 50-1000 rpm, X 4 : 1-40). Here, the flow ratio indicates the ratio of the CO 2 flow rate to the emulsion injection rate (1 ml/min). The process conditions of 35 C, 85 bar, 500 rpm with a flow ratio of 5 were estimated as suitable from the preliminary study, and were set as the center points for the current study. To examine their effect on the response in detail by fine control of process parameters as independent variables within a narrow condition range and to optimize the process based on these conditions, the levels of three process parameters for BBD in this study were set up as presented in Table 1  obtained from ANOVA were considered statistically significant. In contrast, a p-value greater than 0.05 indicated that the model terms were not significant. 93 Further, lack of fit, coefficient of variance (CV, %), coefficient of determination (R 2 ), adjusted coefficient of determination (R 2 adj), predicted coefficient of determination (R 2 pred), adequate precision, and prediction sum of square (PRESS) of the models were also used to determine the adequacy and goodness of fit for each polynomial equation model. 47,[94][95][96] Among the different polynomial models, the best-fit mathematical model was selected based on statistical parameters. The polynomial quadratic equation for the model is given as: where Y is the response; X i -X j are independent variables corresponding to pressure, temperature, stirring rate, and flow ratio, respectively; X i X j and X i 2 are the interaction and quadratic terms, respectively; A 0 is the constant term of the model intercept coefficient (which indicates the overall average response of all runs); and A i , A ii , and A ij is the regression coefficient of the linear, quadratic and interaction terms, respectively. Significant model terms are included in the equation.
To visualize the relationships between responses and independent variables and to deduce the optimal conditions, the fitted quadratic polynomial equation was expressed as a contour plot. The selected best-fitting model was used to optimize the SFEE process for preparing PLGA microspheres with the most desirable target response value. PLGA microspheres were prepared under the predicted optimized SFEE process conditions, and all dependent variables were experimentally obtained.

| Solvent evaporation process
PLGA microspheres were also prepared using the SE process at a condition reported previously for comparison with the SFEE process. 77,78 The emulsion preparation process was identical to that for SFEE. The  T A B L E 2 Box-Behnken experimental design: process parameters as independent variables, measured responses and summary of model fitting and statistical analysis.

Run
Factors Responses

| Brunauer-Emmett-Teller specific surface area
The specific surface area of the prepared blank PLGA microspheres without exenatide was analyzed using TriStar II 3020 (Micromeritics, GA, USA). The sample powders were loaded into standardized sample tubes, and precisely weighed. After completion of degassing procedure using a FlowPrep 060 degasser (Micromeritics, GA, USA), the nitrogen adsorption-desorption isotherm was determined at À196 C. The specific surface area was determined by applying the adsorption-desorption isotherm results to BET theory.

| Water vapor sorption analysis
The moisture sorption/desorption behavior was evaluated using a  Therefore, the amount of residual organic solvent directly analyzed from lyophilized microspheres was used in this study.

| Drug encapsulation efficiency
The sample solution used in CD spectroscopy was also used to determine EE%. The sample solutions were injected into the RP-HPLC system for exenatide quantification, as described above. The EE values were obtained using Equation (2): where LC is the loading capacity % obtained by multiplying the mass of exenatide in the microspheres divided by the mass of the microspheres by 100. A linearly proportional correlation between EE obtained using an indirect method, by determining the amount of drug present in the aqueous phase before freeze-drying, and EE obtained by direct extraction from microspheres was confirmed. Accordingly, the EE value directly analyzed from the microspheres after freezedrying was used in this study.

| In vitro drug release test
The drug release medium was prepared by dissolving tween 20 (0.02% w/v) and sodium azide (0.01% w/v) in pH 7.4 phosphate-buffered saline (PBS), as a dispersing and a preserving agent, respectively. Briefly, 20 mg of PLGA microspheres were accurately weighed and introduced into 2 ml of drug release medium in a glass tube. After starting incubation in a shaking water bath at 37 C and 100 rpm, the medium was collected at predetermined time points and a corresponding volume of medium was newly replenished. The withdrawn medium was collected in microcentrifuge tubes (Eppendorf) and centrifuged (12,000 rpm, 5 C, 10 min) to separate undissolved material; the clear supernatant was the used for exenatide quantification using the RP-HPLC method. The IBR indicates the percentage of exenatide released on the first day. was used to read the absorbance at 450 nm, and the concentration of exenatide was estimated by extrapolation based on a standard curve with good linearity. Direct determinations from the experimentally obtained PK data were used to determine the maximum plasma concentration (C max ) and the time required to reach C max (T max ). In addition, PKSolver software was used for the noncompartmental PK analyses of plasma exenatide concentrations vs. the time profile, and the area under the curve (AUC) was obtained. 103 All data are presented as mean ± SD.

| Pharmacodynamic study to evaluate therapeutic efficacy
Nonfasting glucose levels, food intake, and changes in body weight were measured to investigate in vivo efficacy as PD studies (n = 6-8).
These PD studies for the ELPM_SE and ELPM_SFEE administration group were evaluated simultaneously with the PK study. Additional two groups administered with exenatide aqueous solution (140 μg/3 ml/kg, SC injection twice a day) and an aqueous suspension of blank PLGA microspheres (prepared using SE, single SC injection), respectively, were used as controls to be compared with two groups administered with exenatide-loaded microsphere formulations prepared using SE and the optimized SFEE process. The exenatide aqueous solution and aqueous suspension of empty PLGA microspheres were prepared using the aqueous dispersion medium described above.
For the measurement of nonfasting blood glucose levels, 10 μl of blood samples collected at prescheduled time points were withdrawn using a micropipette and analyzed using a blood glucose meter (AccuCheck Active, Roche Diagnostics, Germany).
Food intake and changes in body weight were monitored thrice a week and once every 3 days, respectively. The average food intake and average body weight change per mouse were determined by dividing the total amount of food intake and total body weight in one cage by the number of animals per cage, respectively.

| Statistical analysis
Significant differences were determined using an independent t-test or one-way ANOVA. All statistical analyses were performed using SPSS v12.0 software (IBM SPSS, Chicago, IL, USA).

| Evaluation of the fitted model and the effects of independent variables on each response
All experimental conditions, where each independent variable was combined using BBD, and the observed responses of particle size, SPAN value, EE, IBR, and residual solvent corresponding to each run are presented in Table 2. The statistical significance of the process parameters (used as independent variables) for the responses was investigated using ANOVA. In the following subsections, the mathematical model explaining the influence of process parameters on each response and their correlation will be discussed in detail.
The model p-value, lack-of-fit, R 2 , R 2 pred, R 2 adj, and adep precision for the four applied mathematical fit models are given in Table 2.
In addition, more detailed statistical analysis results including F-value, coefficient estimates, and PRESS are presented in Table S1 (supple- Generally, R 2 should be at least 0.5 for preferable fitting. 94 In addition, a mathematical model with a better correlation between independent variables and the response has larger values of R 2 and R 2 adj. [104][105][106] Regarding the PRESS value, the smaller the PRESS value, the higher the predictive power, which can prevent the chance of overfitting a mathematical relationship. Based on the model analysis for all the responses in Table 2, the linear and quadratic models were found to have the most desirable R 2 , R 2 adj, and R 2 pred values for particle size and the other four responses (including SPAN value, EE, IBR, and residual solvent), respectively. This indicates that these models are adequate to fit well, with a proper goodness of fit for each corresponding response. Further, the relatively small PRESS value of each selected model indicates a good prediction of the experimental results when compared to that with other models. 107 Consequently, the linear and quadratic models for particle size and the quadratic model for particle size and the other four responses were selected as the most suitable statistical models, X 1 : pressure (bar), X 2 : temperature ( C), X 3 : stirring rate (rpm), and X 4 : flow ratio by applying regression analysis to the experimental data.
The mathematical relationships obtained in the polynomial equations in terms of the actual factors for all the responses are given in Table 2.
Typically, the derived positive or negative signs of the factors in the model equation indicate a positive or negative or synergistic or antagonistic effect on the responses for the factors. [108][109][110] The ANOVA results ( 3.1.1 | Particle size and its distribution As described above, the SR kinetics of drugs from PLGA microspheres can be controlled by particle size and distribution. Thus, the desired drug release profile with uniformity can be obtained by optimizing the PSD. Microspheres that are "too small" exhibit poor SR efficiency; they may also migrate from the site of injection, resulting in undesirable drug release. in contrast, spheres that are "too large" may not easily pass through a syringe needle. 111 Thus, particle size and SPAN value, which is an indicator of PSD, were selected as critical responses for quality control of the SR PLGA microspheres, and were obtained experimentally. 112 From the experimental responses presented in Table 2  To visually evaluate the effects of independent variables on particle size and SPAN value, contour plots are shown in Figure 2a,b. From the plus or minus sign of the terms of actual factors for the response of particle size, it was confirmed that particle size increased as the temperature increased, whereas the particle size decreased as the stir- indicating that an excessive increase in temperature might lead to a decrease in the uniformity of the PSD. 113 In the SFEE process, the droplets of the oil phase undergo swelling after SC-CO 2 diffusion, and undergo shrinking due to DCM diffusion out of the drop into the external water phase. 1,16,114 In contrast, the SC-CO 2 concentration in the drop could be relatively increased.
The emulsion drop behaved as a miniature gas antisolvent precipitator, and the PLGA polymer was fabricated as a microsphere. During this fabrication process, the higher temperature and pressure decreased the shrinkage of the DCM drop, resulting in an increase in the particle size. In particular, the different thermodynamic behaviors in the emulsion could be attributed to the changes in temperature and pressure during the SFEE process, resulting in a difference in the particle size of the microspheres prepared by the SFEE process.
As described above, the particle size, in terms of VMD, of the microspheres increased significantly with increasing temperature. This tion and evaporation, leading to a larger particle size of the obtained PLGA microspheres. This decreased shrinking rate and excessively increased evaporation of DCM can lead to a decrease in the density and more porous internal structure of microspheres ( Figure 3). In addition, the occurrence of particle agglomeration due to an increase in temperature, decrease in stirring rate or flow ratio, which will be discussed later in this subsection, can also be another major reason of the increase in particle size.
However, the insignificant effect of pressure on particle size was Further, as the SFEE process progressed, the CO 2 concentration in the drop increased, and the equilibrium constraints tended to slow CO 2 diffusion into the emulsion droplet until the volume variation became negative (shrinking rate was positive) by comparatively faster diffusion out of DCM from the emulsion droplet. Based on these theories, it was estimated that a significant increase in particle size would not be observed with an increase in pressure within the pressure range used in this study, because the pressure conditions were all less than 100 bar.
On the other hand, it could also be supposed that if the diffusion out of DCM was accelerated by a combination of other factors at a relatively high pressure (not more than 100 bar), the particle size might be decreased with an increase in pressure. However, in this study, pressure had no significant effect on the particle size of the microspheres prepared using the SFEE process. This result may be attributed to that CO 2 diffusion was not limited by equilibrium conditions at 75-95 bar in all experimental runs of this study; thus, an increase in pressure cannot significantly enhance drop shrinkage due to DCM diffusion from the emulsion droplet predominantly more than that of drop swelling because of DCM diffusion into the emulsion droplet.
As stated, the influence of the four process variables on the SPAN value is significant. The increase in SPAN values above 35 C and 85 bar was attributed to particle agglomeration in the suspension, even though the presence of an outer aqueous phase may prevent PLGA microsphere coalescence. 1 The fact that SC-CO 2 can produce The result of this tendency of particle change by stirring rate and flow ratio was also observed in SEM (Figure 3).  The contour plots in Figure 2c,d show the effects of the independent variables on the EE and IBR of ELPMs prepared using the SFEE process. The clearly observed peak values in the counter plots for EE and IBR indicate the significant quadratic effects of pressure, temperature, stirring rate, and flow ratio as the major process parameters.

| EE and IBR
The factor that showed the most pronounced effect on EE was the quadratic term of pressure, and the linear term corresponding to stirring rate showed the most significant effect on IBR.
Interestingly, Figure 2c,d also shows that overall, EE decreased and IBR increased as the process conditions moved away from the point corresponding to the peak value. The estimated process condition corresponding to the peak value was close to the center point at  [119][120][121] Further, the porosity of the fabricated polymer can increase owing to the formation of channels for the mass transfer of the internal phase to the outer phase. [122][123][124][125][126] It was unexpected that the increase in pressure and temperature leaving the dispersed phase during microsphere formation. [127][128][129] Peptide denaturation at high temperatures may also attribute to the decreased EE owing to degradation of the encapsulated peptide. [130][131][132] In addition, the high pressure inside the hydrated PLGA microspheres caused by the combination of high temperature and pressure during SFEE process can induce microsphere breakage. This negative effect could be increased at a temperature higher than the T g of PLGA because the dense outer layer of the microspheres became more relaxable owing to the glass transition of the PLGA polymer. Further, thermodynamic equilibrium exists between the CO 2 dissolved in the aqueous phase and that dissolved in the polymer particles after complete precipitation of PLGA during the SFEE process. Typically, it was reported that between 10% and 30% (w/w%) of CO 2 can be dissolved in the PLGA polymer, depending on the process pressure and temperature. The dissolved SC-CO 2 in the polymer has an effect of "T g lowering" on polymer like PLGA, by which this peptide loss can be increased. 1 The decrease in T g also can be influenced by the processing temperature depending on the characteristics and molecular weight of the polymer, resulting in the formation of CO 2 plasticized polymer particles. 115 This plasticization of the polymer can be beneficial for facilitating faster solvent removal from the polymer phase, but not for forming a hard and rigid polymer layer.
Overall, the IBR value was found to change in the direction opposite to that of the EE for all four process parameters. 133 In general, the IBR of the drug observed in microspheres is probably attributed to poorly entrapped peptides and peptides loosely attached to the internal and outer surfaces. 4 Further, the high porosity of the fabricated polymer, which is strongly correlated with a low EE, may also cause an increase in IBR. 134 Thus, when the EE of the prepared microspheres is low, more drugs are present on the external surface of the microspheres, and can be adsorbed onto the microsphere surface, resulting in a larger IBR. 135-137

| Residual solvent on PLGA microspheres
The experimentally obtained residual solvent for all runs varied in the range of 43.7-243.8 ppm (Table 2).
Three linear terms corresponding to pressure, temperature, and flow ratio and two quadratic terms corresponding to pressure and flow ratio had significant effects on the residual solvent with large coefficients and SME values in terms of coded factors, as well as small p-values (<0.01). Based on this result, we confirmed that the pressure, temperature, and flow ratio are the major process parameters for residual solvents. The order of significance for these effects, determined based on the SME value, was as follows: flow ratio > pressure > temperature > flow ratio 2 > temperature 2 .
The contour plots in Figure 2e show the effects of the process parameters on the residual solvent of ELPMs prepared using the SFEE process. Among significant factors, the linear terms of pressure, temperature, and flow ratio had remarkable effects on the residual solvent, but the effects of quadratic terms corresponding to the pressure and flow ratio were relatively small. Moreover, as shown in the contour plots ( Figure 2e), PLGA microspheres with a lower residual solvent content were obtained as the temperature, pressure, and flow ratio increased.
This result was attributed to the higher mixing efficiency of CO 2 and DCM, and the subsequently improved DCM extraction efficiency. 34

| Determining a desirable range for the combination of various variables in the SFEE process by graphical optimization
Based on the above experimental results within the design space applied in this study, optimization was carried out to verify the fitted model after establishing polynomial equations to describe the relationship between the factors and the processed responses by setting the acceptable target ranges of responses, as given in Table 1.
The target range of particle size was set to 20-30 μm because it was reported that the mean particle size of Bydur-eon™ microspheres is approximately 50 μm, which requires the use of painfully large, 23-gauge needles for SC injection, thus decreasing patient compliance. [56][57][58] The SR exenatide delivery system including Bydureon™ or Byetta ® Long-Acting Release (LAR) was developed by Amylin Pharmaceuticals/Eli Lilly using Alkermes's water-in-oil-in-oil (W/O/O) coacervation technology. 55 Based on the results of our previous and preliminary studies, the target range of the SPAN value was selected as 1.3 or less. The target range of EE was set at a minimum of 85%, which could be easily achieved through SFEE, with an ideal of 100% as the maximum. Based on the results of our previous studies and the in vivo evaluation reports on the IBR of the exenatide long-acting formulation, the target range of the IBR was set to a minimum of 3.3% and a maximum of 10%, assuming a SR formulation for 30 days. The ICH guidelines stipulate that DCM be less than 600 ppm, but to apply a stricter management standard, 150 ppm, 1/4th of the standard set by ICH guideline, was set as the maximum target range. 58 The 4D sweet-spot plots shown in Figure 4 were obtained using the four factors and an acceptable range of responses, as given in Table 1. The model error and process uncertainty were not considered because a deterministic regression model was applied for the construction of this sweet-spot plot. The yellow area is the sweet spot, which clearly shows the range of conditions for combinations of different process parameters that meet the criteria for all responses. The condition associated with the center point was thus confirmed to be suitable, as it was located within the yellow area covering all target response ranges. Thus, the ELPMs prepared using the SFEE process at 35 C, 85 bar, 500 rpm stirring rate, and a flow rate of 5, were then used for evaluating the following results.
As such, a fundamental understanding of the effects of the SFEE process parameters on the pharmaceutical properties of ELPM and F I G U R E 4 Four-dimensional sweet-spot plots obtained by plotting the factors (X1-X4) leading to optimal responses information on the range of suitable process conditions that can derive the desirable pharmaceutical quality of SR formulation can be applied to commercially scale-up manufacturing of PLGA microsphere as translational research for successful commercial production and clinical use.

| Comparison of PLGA microspheres prepared using SFEE and solvent evaporation
As mentioned above, PLGA microspheres obtained from the center point of the design space that satisfy all response criteria were compared with PLGA microspheres prepared using a previously reported SE method through various solid-state characterizations and in vitro and in vivo evaluations (Table 3 and Figure 5).

| PSD and morphology
The particle size of the ELPMs prepared using the SFEE process (ELPM_SFEE) was smaller than that prepared using the SE process (ELPM_SE) ( Table 3). This may be attributed to different extraction mechanisms with volume expansion and shrinking rates during extraction. In addition, the smaller SPAN values of ELPM_SFEE than ELPM_SE shows a more uniform PSD (Table 3). The SEM images in Figure 5a indicated that the morphology and surface properties of ELPM_SFEE were more spherical and smoother than those prepared using SE. Many previous studies have reported that polydispersed PLGA microspheres can be generated using conventional microsphere preparation techniques such as SE, and must be filtered or sieved to isolate particles within the desired particle size range. 3 This result may be attributed to the higher extraction efficiency of the SFEE process than that of the SE process. In particular, the fast solidification of the dense outer PLGA layer using the SFEE process may prevent the easy release of the inner water phase during the extraction process, allowing the preparation of a spherical microsphere with a smooth surface. 16 Moreover, the mean particle size of ELPM_SFEE was approximately 20 μm, which was much smaller than those of ELPM_SE and Byetta ® LAR microspheres (i.e., approximately 50 μm). Byetta ® LAR requires the use of painfully large, 23-gauge needles for SC injection and, thus, decreases patient compliance. 59 In contrast, an injectability test confirmed that only the suspension of ELPM_SFEE for SC injection can be injected through a 25-gauge needle. Thus, ELPM_SFEE with a smaller particle size could improve patient compliance.

| Residual organic solvent
Comparatively, the extraction of DCM from the same volume of emulsion using the SE process at 35 C required approximately 7 h to reach DCM levels below 600 ppm. In contrast, with the SFEE method, 5 h was sufficient for the level of residual DCM to fall below 150 ppm ( Table 3). As mentioned above, these results are attributed to the higher mixing efficiency of CO 2 and DCM, and the subsequent improved DCM extraction efficiency in the SFEE process than that in the SE process. Therefore, from these residual solvent analysis results, it was confirmed that the SFEE technology is not only a production technology for SR microspheres that can efficiently minimize the presence of organic solvents in a short process time, but also have the potential clinical benefits which can prevent the increase in the risk of toxicity in patients caused by residual organic solvents.

| BET specific surface area and water vapor sorption
The blank PLGA microspheres without exenatide were prepared under the optimized SFEE process conditions as described above (at center point), and used as a substitute for material sparing in BET measurements requiring large amounts of sample. Blank microspheres were also prepared using SE for comparison. In both processes, the presence or absence of exenatide did not significantly affect the particle size of the microspheres.
T A B L E 3 In vitro characteristics and in vivo pharmacokinetic parameters of exenatide following a single SC injection to streptozotocin-induced diabetic mice of ELPM_SE and ELPM_SFEE. F I G U R E 5 Comparison of (a) scanning electron micrographs (SEM), (b) water vapor sorption profiles over increasing relative humidity (RH%), and (c) in vitro exenatide release (n = 3, pH 7.4 PBS, 37 C) between exenatide loaded PLGA microsphere prepared by SE (ELPM_SE) and SFE process (ELPM_SFEE, (d) circular dichroism spectra with change in the conformation type as determined using the Bestsel secondary structure analysis tool and (e) second derivative Fourier-transform infrared (FT-IR) spectra for comparison of secondary structure between raw exenatide and extracted exenatide from ELPM_SE and ELPM_SFEE, (f) pharmacokinetic profiles and (g) changes in nonfasting blood glucose levels, (h) food intake, and (i) weight gain in streptozotocin-induced diabetic mice: a single SC injection with ELPM_SE or ELPM_SFEE, twice daily SC injection with exenatide solution, or a single SC injection with blank PLGA microsphere. ("*" indicates significant difference (P < 0.05) compared to ELPM_SE, "+" indicates significant difference (P < 0.05) compared to exenatide solution).
As presented in Table 3, the surface area and pore size of blank microsphere prepared by the SFEE process were significantly smaller than those prepared using the SE process. This result was also confirmed by the water vapor sorption results (Figure 5b) indicating that the degree of water vapor sorption of ELPM_SFEE can be significantly lower than that of ELPM_SE. The water channels connecting the internal and outer water phases can be formed during PLGA fabrication using both evaporation and extraction processes resulting in pores in the microspheres. However, in the SE process, these channels in PLGA microspheres can be formed excessively owing to the generation of vaporized gas. This difference is presumed to have led to the observed results. This also indicates that the stability of ELPMs prepared using the SFEE process during storage could be improved by lowering the decomposition rate of peptides owing to the low moisture content. These results also support the PSD and morphology results described above. In particular, some small pores were rarely observed in the SEM images of ELPM_SE but none were observed for ELPM_SFEE.

| EE, IBR, and sustained release profile
The EE of the SFEE microspheres containing exenatide was higher than that of SE microspheres (Table 3) because PLGA microspheres prepared using the SFEE process solidify rapidly, forming a dense thin outer layer of microspheres due to the high efficiency of diffusion and extraction rate of SC-CO 2 when compared with the SE process. These results indicate that the outer layer of PLGA microspheres prepared using the SFEE process was denser and thicker than that of the SE-processed PLGA microspheres. Naturally, a dense outer layer provides impedance, hindering peptide diffusion through the polymer layer during solvent removal. [138][139][140] Figure 5c shows the cumulative release of exenatide from ELPM in vitro. A typical triphasic release pattern was observed for both ELPM_SE and ELPM_SFEE, with IBR followed by a more controlled secondary phase and accelerated drug release in the third phase. 54 As expected, the IBR of the SFEE microspheres with higher EE values was lower than that of the SE microspheres (Table 3).
Low EE and high IBR of ELPM_SE mean that exenatide molecules exiting the outer water phase leaked from inner water phase through the water channels are more adsorbed on the surface of ELPM_SE than ELPM_SFEE. Excessive formation of these water channels could be attributed to the larger pore size of the PLGA microspheres prepared using SE, compared to SFEE, hence resulting in the increased IBR of ELPM_SE by the larger surface area due to the pores of microsphere. 141 In addition, a polymer impregnated with CO 2 at a higher pressure, within a certain range of pressure and temperature, has been reported to show relatively low porosity in many research cases. 138,[142][143][144] Based on this hypothesis, the BET specific surface area (Table 3) and water vapor sorption results ( Figure 5b) can explain why the EE and IBR of the microspheres prepared using this process were smaller than those with the SE process.
The amount of exenatide released from ELPM_SE was less than 10% between day 7 and 14 in the secondary phase, which indicate it has obvious lag time, which means release of a drug that is inappropriately slow (Figure 5c). In addition, the final accumulated release per- Consequently, the SFEE process was revealed to produce ELPMs with a more desirable SR property than that of PLGA microspheres produced using the commercially used conventional SE process.

| Secondary structure of exenatide in PLGA microspheres
Circular dichroism spectroscopy was used to evaluate the secondary structure of exenatide in aqueous solutions. Exenatide in the fabricated microspheres was extracted and used for CD analysis along with raw exenatide. As reported in our previous studies, the CD spectra of exenatide had two specific minima at wavelengths of 208 and 222 nm, indicating the presence of an α-helix (Figure 5d). 71,74 The obtained CD spectroscopic data were also analyzed using the BeStSel analysis tool to determine secondary structures, such as α-helices, β-sheets, β-turns, and other structures. 146 Further, second-derivative FT-IR spectra were obtained for unprocessed exenatide and exenatide in microspheres prepared using the SE and SFEE processes (Figure 5e) to check for differences in secondary structure between samples.
Intermolecular β-sheets could be identified by the two peaks at 1692 and 1629 cm À1 , and the peak assigned to the α-helix could be observed at 1660 cm À1 in the amide I region. 66,76,97 Typically, meaningful information about the peptide secondary structure could be obtained from these representative peaks in the second derivative FT-IR spectra. There was no significant difference between the extracted exenatide from PLGA microspheres prepared using the SFEE and SE processes when compared to the unprocessed exenatide, indicating that the exenatide secondary structure in SE-and SFEE-processed microspheres can be maintained without structural destabilization even after the completion of all manufacturing processes. 69,77 These results are presumed to be attributed to the stabilization of exenatide by the formulation containing various stabilizers and additives, which can ensure structural stability, as found in our previous study.
3.4 | In vivo animal study in STZ-induced diabetic mice

| PK profiles
The comparative PK profiles of ELPMs prepared using SE (ELPM_SE) and SFEE (ELPM_SFEE) and blank PLGA microspheres after a single SC injection are shown in Figure 5f. The calculated PK parameters are given in Table 3.
The curve in the PK profiles shows double peaks, and not a single peak, for both microspheres, typically indicating the IBR and subsequent SR of exenatide from PLGA microspheres. 55 Specifically, the blood concentration of exenatide rapidly increased and immediately decreased within the first day after SC administration, which was observed in both ELPM_SE and ELPM_SFEE. Further, there was a clear difference in the degree of IBR between the two formulations.
The in vivo IBR of exenatide is suggested to lead to the appearance of the first peak of exenatide plasma concentration in the PK profile. 74 As presented in Table 3 tained PK profile compared with that of Byetta ® LAR, a commercial ELPM product, based on a report that it has poor SR characteristics with low release between days 4 and 17 in animal PK study. 55 Like this, it has been reported that Byetta ® LAR, comprising SR exenatide PLGA microspheres, has some problems. In particular, the weekly dose of exenatide in Byetta ® LAR microspheres (i.e., single injection of 2 mg/human) is much higher (14-28 fold) than that of Byetta ® (i.e., twice-daily injection of 70-140 ug/human). The improper release of exenatide from Byetta ® LAR in vivo may explain its poor pharmacokinetics and the necessity for a much higher dose. 60 Another study reported that the drug burst release was high (approximately 45%) when they prepared microspheres following the commercialized ELPM formulation. 59 These problems can increase the risk of adverse effects due to overdose of exenatide. 54

| Pharmacodynamic therapeutic efficacy
The measured nonfasting glucose levels of the ELPM_SE, ELPM_SFEE, and blank PLGA microspheres after a single SC injection and twice daily SC injection of exenatide solution are shown in Figure 5g. For the empty PLGA microsphere-injected group, nonfasting glucose concentrations increased progressively over 32 days, indicating no blood glucoselowering effect in STZ-induced diabetic mice. Twice daily SC administration of exenatide solution had a slightly higher hypoglycemic efficacy compared with that of empty microsphere administration. In contrast, significantly reduced nonfasting glucose levels were observed in the two groups administered a single SC injection of ELPM_SE and ELPM_SFEE when compared with the twice daily SC injection of exenatide solution.
In particular, ELPM_SFEE lowered nonfasting blood glucose concentrations more effectively than ELPM_SE, and this difference in hypoglycemic efficacy was significant. Overall, ELPM_SFEE, which has a lower IBR, higher EE, and sustained in vitro and in vivo release, showed proper in vivo therapeutic efficacy for hypoglycemic behavior. 54 Furthermore, as shown in Figure 5h,i, administration of exenatide solution (twice daily SC), ELPM_SE (single SC), and ELPM_SFEE (single SC) significantly decreased food intake and weight gain in STZ-induced diabetic mice compared to the control group administered empty PLGA microspheres.
The reduction in food intake and weight gain was the greatest in the ELPM_SFEE group, and the order of these two PD effects was as follows: ELPM_SFEE > ELPM_SE > exenatide solution > blank PLGA microspheres.
Based on the results of PD therapeutic efficacy studies, ELPM_S-FEE had the most desirable in vivo efficacy for decreasing blood glucose levels, weight gain, and food intake. 52

| CONCLUSION
In this study, a BBD with four factors and three levels was applied to investigate the effect of process parameters on the fabrication of ELPMs by SFEE. The four process parameters selected as independent variables were pressure (X 1 ), temperature (X 2 ), stirring rate (X 3 ), and flow ratio (X 4 ), and the effects of independent variables on five responses, including particle size and its distribution (SPAN