Enhancing the Poor Flow and Tableting Problems of High Drug-Loading Formulation of Canagliflozin Using Continuous Green Granulation Process and Design-of-Experiment Approach

Maximization of drug-loading can significantly reduce the size of dosage form and consequently decrease the cost of manufacture. In this research, two challenges were addressed: poor flow and tableting problems of high-drug loading (>70%) formulation of canagliflozin (CNG), by adopting the moisture-activated dry granulation (MADG) process. In this method, heating and drying steps were omitted so, called green granulation process. A 32 full-factorial design was performed for optimization of key process variables, namely the granulation fluid level (X1) and the wet massing time (X2). Granulation of CNG was carried out in the presence of polyvinylpyrrolidone, and the prepared granules were compressed into tablets. Regression analysis demonstrated the significant (p ≤ 0.05) effect of X1 and X2 on properties of granules and corresponding tablets, with pronounced impact of X1. Additionally, marked improvement of granules’ properties and tableting of CNG were observed. Furthermore, the optimized process conditions that produced good flow properties of granules and acceptable tablets were high level of granulation fluid (3.41% w/w) and short wet massing time (1.0 min). Finally, the MADG process gives the opportunity to ameliorate the poor flow and tableting problems of CNG with lower amounts of excipients, which are important for successful development of uniform dosage unit.


Introduction
In 2013, the U.S. Food and Drug Administration had approved Canagliflozin (CNG) for the management of adult patients with type-II diabetes mellitus [1]. The CNG is one of the orally acting sodium-glucose co-transporter-2 inhibitors that reduces the renal tubular reabsorption of glucose into the mathematical relationship between the independent variables and dependent response(s) [15,16]. The significance of DoE as an integral part of the drug product development has been reported [6,17]. Development of high-loading tablet formulation of CNG using MADG has not been reported in the literature.
Based on these premises, the aim of the current investigation was to (1) address the poor flow and tableting problems of CNG using the continuous MADG method, (2) application of the DoE approach to evaluate the effect of key process variables of MADG and their interactions on critical quality attributes (CQAs) of intermediate granules and corresponding tablets and (3) optimization of the key process variables using the desirability function to provide CNG tablets with desired attributes.

Statistical and Diagnostic Analysis of the Design
Multiple regression analysis results of the proposed models are given in Table 1. Design-expert software can generate mathematical polynomial models (i.e., linear, 2-factor interaction, quadratic and cubic) to relate the variables to the responses. For each response, p < 0.05 indicates that the terms in the model reflect the behavior of the response function. In addition, the adjusted R 2 of the selected model reasonably agreed with the corresponding predicted R 2 , with determination coefficients (R 2 ) more than 0.8011, confirming the convenience and accuracy of the selected models [18]. Moreover, diagnostic plots were generated for granules and tablet responses to evaluate the goodness of fit of the applied model and confirm its significance. Linear correlation plots (Figure 1) between the actual and the predicted values with higher R 2 indicate good model fit.

Mean Granules Size (d 50 )
As shown in Table 2, increasing the amount of granulation fluid from 1% to 4% w/w and the wet massing time from 1 to 5 min led to an increase of d 50 from 117.12 ± 0.25 to 379.14 ± 0.33 µm, a decline in the percent fines (<50 µm) from 28.08% ± 0.013% to 2.01% ± 0.024% and a decrease in the distribution width (span) from 3.64 ± 0.013 to 1.26 ± 0.04, which are evident for granules formation by the MADG process. Analysis of variance (ANOVA) ( Table 3) revealed that X 1 and X 2 had a significant effect on granules' d 50 (p < 0.0001 and p = 0.0002, respectively), with a pronounced impact of X 1 , as indicated by its high sum of squares (45,562.02 for X 1 and 7816.37 for X 2 ). In addition, the d 50 was positively correlated with X 1 and X 2 , as evident by the sign of their coefficient estimates (+87.14 for X 1 and +36.09 for X 2 ). Figure 2 shows that the d 50 significantly increases with increasing both X 1 and X 2 . Otherwise, X 1 and X 2 had a significant (p < 0.0001 for X 1 and p = 0.0009 for X 2 ) impact on the percent fines, with the highest impact of X 1 , as demonstrated by its higher sum of squares (611.45 for X 1 and 79.28 for X 2 ). Furthermore, values of coefficient of estimation indicated that X 1 had a marked effect on percent fines in a negative direction, while X 2 with a low coefficient of estimation value had a low impact in the same direction (−10.10 for X 1 and −3.36 for X 2 ), as shown in Figure 2. This suggests that granulation with a high amount of granulation fluid and long massing time results in producing a low level of fines. Increment of the granulation fluid results in adequate wetting of the particles' surfaces and buildup of liquid bridges among the particles that improve the granules' coalescence and growth [13]. In addition, increasing the wet massing time leads to increasing the frequency of particles collision, which improves the granule growth and reduces the percent fines [19]. It was reported that granule growth had been influenced by the mechanical shear of the MADG process [20]. Figure 1. Linear correlation plot relating mean granule size, percent fines, distribution width, bulk density, angle of repose, SD of weight variation, breaking force, friability, disintegration time and drug release at 30 min, between the predicted and the measured (actual) values. SD: standard deviation. The distribution width (span) determines the broadness of granule size distribution and has a significant effect on granules' flow, compressibility and segregation [21]. A high value of distribution width indicates a wide size distribution of the system, and vice versa. As shown in Table 3, X 1 and X 2 were found to be statistically significant (p < 0.0001 and p = 0.0008, respectively) with respect to their effect on distribution width, with a prominent impact of X 1 , as evident by its higher sum of squares value (4.05 for X 1 and 0.6534 for X 2 ). The distribution width was found to be negatively correlated with X 1 and X 2 according to the sign of their coefficient estimates (−0.8217 for X 1 and −0.330 for X 2 ). As shown in Figure 2, the distribution width was decreased as the granulation fluid and the wet massing time increased. The generated regression equations that demonstrate the influence of X 1 and X 2 on mean granule size, percent fines and distribution width, in terms of coded variables, were expressed as follows: Mean granule size (µm) = 254.39 + 87.14 × X 1 + 36.09 × X 2 (1) Percent fines (%) = 16.13 − 10.10 × X 1 − 3.36 × X 2 (2)

Granules' Bulk Density
As indicated in Table 2, increasing the granulation fluid and the wet massing time resulted in increasing the granules' density from 0.353 ± 0.013 to 0.401 ± 0.006 gcm −3 . Regression analysis (Table 3) showed that X 1 and X 2 had a significant (p = 0.0024 and p = 0.0007, respectively) impact on granules' bulk density. However, the influence of X 2 on bulk density was more pronounced than that of X 1 according to sum of squares values (0.0005 for X 1 and 0.0009 for X 2 ). Granulation with long wet massing time resulted in exposing the prepared granules to a high shear force for a long period, that led to a decrease of the porosity of granules as well as an increase of the granules' consolidation and density [22]. The effects of studied variables on granules' density are shown in Figure 2. The X 2 had a higher effect on bulk density in a positive direction, while the X 1 had a low impact in the same direction, as evident by the sign of parameter estimates (+0.0092 for X 1 and +0.0120 for X 2 ). On the other hand, the two-way interaction between X 1 and X 2 also had a significant (p = 0.0252) impact on bulk density of the prepared granules.
The generated regression equation that demonstrates the influence of X 1 and X 2 on bulk density of granules in terms of coded variables was expressed as follows:

Granules' Flow
As displayed in Table 2, the angle of repose decreased from 32.11 ± 0.322 • to 26.23 ± 0.415 as the amount of granulation fluid and the wet massing time increased, demonstrating an improvement in powder flow upon granulation using the MADG method. Moravkar et al. reported that the MADG process could produce excellent flowability granules of high drug-loading formulation because of the uniform size of the prepared granules [10]. The results of ANOVA analysis (Table 3) demonstrated that X 1 and X 2 had a significant (p < 0.0001 and p = 0.0002, respectively) negative effect on the angle of repose of prepared granules, as evident by the negative sign of their coefficient estimates (−2.12 for X 1 and −0.9317 for X 2 ). Nevertheless, X 1 was the predominant variable, as evident by its higher sum of squares (26.92 for X 1 and 5.21 for X 2 ). Further, the two-way interaction between the two variables had a significant (p = 0.0356) negative impact on granules' flow. Figure 2 demonstrates the inverse correlation of tested variables on the angle of repose. As explained above, the MADG process at high amounts of granulation fluid and long massing time resulted in an increased size of granule and bulk density, and a reduced amount of fines. This led to a reduced angle of repose and improved the flow of obtained granules [23].
The generated regression equation that demonstrates the influence of X 1 and X 2 on angle of repose of granules in terms of coded variables was expressed as follows:

Tablet Weight Variation
The main cause for granulation is to provide tablets with acceptable weight variation to assure drug content uniformity, which depends on powder flow [24]. For all runs, average tablet weight and its SD are presented in Table 4. With respect to United States Pharmacopeia (USP) criteria, the variation of tablet weight was acceptable for all runs and the SD was less than 1.9, demonstrating proper granules flow. However, little variation observed in the tablets' weight was due to variation in granules' bulk density, as previously discussed in Section 2.2.2. Regression analysis (Table 5) revealed that X 1 and X 2 had a significant (p < 0.0001 and p = 0.004, respectively) negative impact on the SD of tablet weight variation, as indicated from their values of coefficient estimates (−0.1633 for X 1 and −0.0733 for X 2 ). However, X 1 has a dominant effect due to its higher sum of squares (0.1601 for X 1 and 0.0323 for X 2 ). As shown in Figure 2, both variables have a negative correlation with the SD, indicating that an increase in each variable individually results in a reduction of the SD. In addition, the low SD value was observed in granules prepared at a high level of granulation fluid and long wet massing time, as shown at the higher right corner of the contour plot. This is due to the amelioration of flow of elaborated granules upon increasing the level of granulation fluid and the wet massing time, as previously discussed in Section 2.2.3. Table 4. Physical properties of prepared canagliflozin tablets (mean ± SD).

Runs
Weight (mg ± SD) The resulted regression equation that demonstrates the effect of X 1 and X 2 on SD of tablet weight variation in terms of coded factors was as follows: SD of tablet weight variation = 1.66 − 0.1633 × X 1 − 0.0733 × X 2

Tablet Breaking Force and Friability
As the tablet strength is significantly related to the drug release in the patient's body, it is essential to determine the tablet mechanical strength (i.e., breaking force and friability) [25]. As shown in Table 4, the tablet breaking force decreased from 7.92 ± 0.65 to 6.22 ± 0.23 KP upon the granulation fluid, and the wet massing time increased. ANOVA analysis (Table 5) revealed that X 1 and X 2 had a significant (p = 0.0004 and p = 0.0008, respectively) effect, but nearly equal on tablet breaking force, as evident from values of their sum of squares (1.42 for X 1 and 1.05 for X 2 ). In addition, change in X 1 and X 2 had almost the same effect on tablet breaking force in a negative direction, as indicated by the negative sign of their coefficient estimates (−0.4867 for X 1 and −0.4183 for X 2 ). Figure 2 shows that granules prepared at a high level of granulation fluid and long wet massing time produce tablets with a small breaking force. As previously discussed in Section 2.2.2, granulation with a high level of granulation fluid and long wet massing time produced less porous and denser granules with a low fragmentation tendency, which resulted in low tablet breaking force, and vice versa [26]. It was reported that the densified granules prepared by the MADG method might hinder fragmentation, especially when the amount of granulating fluid (water) exceeds 3% [20]. The low compressibility of less porous and high-density granules had been previously reported [22,27].
As shown in Table 4, runs with low levels of granulation fluid (runs 1, 2 and 3) produced friable tablets (i.e., friability > 1.0%). This phenomenon is due to inappropriate wetting of powder particles and formation of high amounts of fines [28]. This finding obviously indicated the inadequacy of these granulation runs for preparing reasonable tablets. On the other hand, granulation runs with high levels of granulation fluid (runs 4-9) produced acceptable tablets with respect to the USP limit (i.e., friability < 1.0%). ANOVA analysis (Table 5) showed that only X 1 had a significant (p < 0.0001 for X 1 and p = 0.2955 for X 2 ) effect on friability of the tablet in a negative direction, as evident by the negative sign of coefficient estimates (−0.335 for X 1 and −0.0367 for X 2 ). Figure 2 demonstrates that an increase in the level of granulation fluid leads to a decrease of friability for tablets prepared by the MADG method.

Tablet Disintegration
As shown in Table 4, disintegration time was found to be increased from 9.34 ± 1.75 to 20.29 ± 0.89 min upon the level of granulation fluid, and the wet massing time increased. ANOVA analysis (Table 5) indicated that X 1 and X 2 had a significant (p < 0.0001 and p = 0.0022, respectively) effect on tablet disintegration time in a positive direction, as evident by the positive sign of their coefficient estimates (+4.49 for X 1 and +1.15 for X 2 ). In addition, the two-way interaction of two variables had a significant (p = 0.0189) effect in a positive direction (coefficient of estimate = +0.67). Nevertheless, X 1 had a pronounced impact on disintegration time with respect to the value of its sum of squares (120.87 for X 1 , 7.96 for X 2 and 1.8 for X 1 X 2 ). Figure 2 indicated that the prolonged disintegration time was associated with a combination of a high level of granulation fluid and long wet massing time, as displayed in the higher right corner of the contour plot. This finding could be attributed to the fact that granulation at high levels of granulation fluid and long wet massing time produce less porous and high-density granules, as previously discussed in Section 2.2.2. These granules retard the penetration of water inside the tablets and delay the disintegration time [28]. This result is in agreement with that previously reported by Takasaki et al. [20]. They reported that the granules prepared by MADG were denser and harder, which hinder fragmentation of the granules and penetration of water into the tablets. Further, a combination of a high amount of granulating fluid and long wet massing time reduced the amounts of fines, which are essential for tablets' disintegration [29]. Takasaki et al. reported that increasing the water amount delayed the disintegration time of tablets prepared by the MADG process [12].

Tablet Dissolution
Tablet dissolution is a key factor as it controls the drug release from the tablet, and therefore its bioavailability [30]. Release profiles of CNG tablets are presented in Figure 3. According to the USP standards for immediate release tablets, all prepared tablets showed acceptable drug release (i.e., 85% release within 30 min) except runs 7, 8 and 9 which released 82.32% ± 3.97%, 78.35% ± 4.61% and 74.21% ± 4.15% respectively, after 30 min (Table 4 and Figure 3). Regression analysis (Table 5) revealed that X 1 and X 2 had a significant (p = 0.0002 and p = 0.007, respectively) negative effect on drug release, as evident by the negative sign of their coefficient of estimates (−5.86 for X 1 and −2.81 for X 2 ). However, X 1 was the most important variable, as shown by its higher sum of squares compared to X 2 (206.27 for X 1 and 47.38 for X 2 ). Figure 2 showed that both variables were inversely proportional with the percentage of drug release after 30 min. In addition, formulations prepared at a combination of a low level of granulation fluid and short wet massing time showed faster release than others. This could be attributed to the fact that granulation at low levels of granulation fluid and short wet massing time produce small size and low-density granules, as previously described, which rapidly eroded, disintegrated and rapidly released the drug [27]. The resulted regression equation that demonstrates the effect of X 1 and X 2 on percentage of drug release after 30 min in terms of coded factors was expressed as follows: Drug release after 30 min (%) = 84.96 − 5.86 × X 1 − 2.81 × X 2 (10)

Optimization of Process Variables Using Desirability Function
The main purpose of the optimization step was to optimize the process variables to produce products with desired properties [25]. As shown in Table 6, numerical optimization using the desirability function was done by setting goals for each dependent response. For successful CNG tablet formulation, acceptable weight variation, mechanical strength, disintegration time and percent release with respect to USP standards are required. It was expected that the independent variables that produce CNG tablets that complied with the USP standards would be 3.41% w/w and 1.0 min for granulation fluid and wet massing time respectively, with an overall desirability of 0.818 ( Figure 4). As presented in Table 7, the observed values of breaking force (7.15 ± 1.83 KP), friability (0.71 ± 0.95 %), disintegration time (13.56 ± 0.76 min) and drug release at 30 min (87.25 ± 2.13 %) were in close agreement with the predicted values of breaking force (7.39 KP), friability (0.69%), disintegration time (14.0 min) and drug release at 30 min (84.21%). In addition, the small value of calculated relative errors (<5.0%) assured the validity of the applied design.

Experimental Design
Before generation of the design, preliminary studies were performed to determine the independent variables and ranges of each variable at which proper granules and tablets were obtained. The 3 2 full-factorial design was done to explore the influence of two process variables: granulation fluid level (X 1 ) and wet massing time (X 2 ), on the critical quality attributes (CQAs) of granules and tablets using Design-Expert ® software Version-11 (State-ease, Inc. Minneapolis, USA). Each variable was evaluated at three levels, coded as −1, 0 and +1, for low, medium and high, respectively ( Table 8). The full matrix of the generated design is shown in Table 9. The run at the center point was performed in triplicate to validate the design and prevent the experimental error. The ANOVA test using Design-Expert software was applied to investigate the influence of independent variables on the studied dependent responses at the 95% level of significance. In order to suggest the significance of the selected model, the R 2 and p-value (should be <0.05) of the proposed models were compared. The general polynomial equation applied for the 3 2 factorial design is as follows: where β 0 is the arithmetic mean response of all runs, and β 1 , β 2 , β 3 , β 4 and β 5 are regression coefficients of estimate of the independent variables X 1 and X 2 . X 1 X 2 and X 1 2 and X 2 2 represent the interaction and quadratic effect, respectively. The relative error was determined using the following formula [31]:

Manufacture of Granules and Tablets
The formulation used in the current study was shown in Table 10. MADG runs (the batch size was 400 g) were performed in a high-shear mixer/granulator (BOSCH Packaging Technology, Schopfheim, Germany). The specified amount of CNG and polyvinylpyrrolidone were dry-blended in the high-shear mixer for two minutes at high speeds of impeller and chopper (300 and 2000 rpm, respectively). The produced blend was then granulated by addition of a very small amount of granulation fluid (1-4% w/w) using a binary spray nozzle. Distilled water was used as granulation fluid in the current study. Following water addition, the blend was wet massed for a particular wet massing time according to the experimental design. For the two min absorption stage, the chopper was stopped and absorbent materials (i.e., microcrystalline cellulose and colloidal silicon dioxide) were added. Finally, disintegrant croscarmellose sodium and pre-sieved lubricant magnesium stearate were directly added to the blend and mixed in the granulator for 2.0 min and 1.0 min respectively, at low impeller speed (200 rpm). The final blend was then compressed into 400 mg tablets at a compression force of 13 KN using a rotary tablet press (RoTap-T 2.0, Kg pharma, Berlin, Germany). The machine was adjusted to provide four tablets per run using 10 mm flat tablet tooling. The obtained tablets were collected for further investigation. Mean granule size of the prepared granules was measured using the dry dispersion technique of the laser diffraction particle size analyzer (Mastersizer 2000, Malvern Instruments Ltd., Worcestershire, UK). Approximately 5-6 g samples were air-dispersed at an inlet air pressure of one bar and a feed-rate of 30%. Obscuration was adjusted between 0.6% and 6%.

Granules' Bulk Density
Bulk density (ρ b ) was measured by carefully pouring the granules into a 50 cm 3 graduated cylinder. Bulk volume (Vb) of the granules sample and corresponding mass (M) were determined. The ρ b was determined using Equation (13) [32]:

Granules Flow
The granules flow was determined using the static angle of repose procedure. A dry funnel was clamped at 2 cm (H) above a clean paper placed on a flat surface. The granules were carefully poured through the dry funnel until the apex of the cone, just reaching the tip of the funnel. The average diameters (D) of the cone base were determined and the angle of repose was determined using Equation (14) [32]:

Weight Variation
Individual weight of twenty randomly selected tablets was determined using an analytical balance (Mettler Toledo New Classic ML204/01, Ohio, OH, USA). The weight variation was assessed by considering the standard deviation (SD) of tablet weight. Results are presented as mean ± SD.

Breaking Force
Individual breaking force (hardness) for ten randomly selected tablets was measured using an automatic hardness tester (Erweka Multi-Check 5.1, Heusenstamm, Germany). Results are presented as mean ± SD.

Friability
Friability of prepared tablets was done according to the method mentioned in USP [32]. Ten randomly selected tablets were accurately weighed (W 1 ) using an analytical balance (Mettler Toledo New Classic ML204/01, Ohio, OH, USA) and placed in a friability tester (Toyama Sangyo TFT-1200, Osaka, Japan), rotated at 25 rpm for 4 min. The tablets were de-dusted and accurately weighed (W 2 ). Friability was calculated using Equation (15):

Tablet Disintegration
Disintegration of prepared tablets was carried out according to the method described in USP [32]. The in vitro disintegration test for six randomly selected tablets was done using a disintegration tester (Erweka, ZT4, Heusentsamn, Germany) in 800 mL of distilled water kept at 37 ± 0.5 • C. For each tablet, disintegration time was recorded in minutes when all solids passed through the screen of the disintegration apparatus. Results are presented as mean ± SD.

Tablet Dissolution
The drug release for six randomly selected tablets was performed using a dissolution tester (Distek 2500, Distek Inc., New Jersey, NJ, USA) following the USP paddle method [32]. The test was done in 900 mL of pH 6.8 phosphate buffer solution (with 0.75% w/v sodium lauryl sulphate) kept at 37 ± 0.5 • C with a paddle rotation speed at 75 rpm. After specified time intervals of 5, 10, 15, 20, 30 and 45 min, samples of dissolution medium were analyzed using in situ fiber optic UV testing (Distek Opt-Dis 410, Distek Inc., New Jersey, NJ, USA) at λ max of 290 nm [5,33].

Conclusions
MADG showed good opportunities to address the problems of poor flow and tableting problems of canagliflozin, which is critical for effective development of small size tablets of high-loading drugs, and consequently, improved the patient compliance due to ease of tablet swallowing. Quantitative correlation between MADG key process variables, intermediate granules and final product tablets have been established using the design of experiment approach. In particular, regression analysis of obtained data demonstrated the significant (p ≤ 0.05) effect of tested process variables on properties of granules and corresponding tablets, with a pronounced impact of the granulation fluid level. The levels of optimized process conditions that produce good flow granules and acceptable tablets were high level of granulation fluid (3.41% w/w) and short wet massing time (1.0 min). From an industrial perspective, application of the MADG technique can significantly decrease the manufacturing cost, as obtained granules could be directly compressed into tablets without drying and milling. In summary, this study increased knowledge of the influence of key process variables of MADG on properties of granules and tablets and could serve as a backbone for further developing a mechanistic model for the MADG process.