## Abstract

Curcumin has been found to possess significant pharmaceutical activities. However, owing to its low bioavailability, there is a limitation of employing it towards clinical application. In an attempt to surmount this implication, often the choice is designing novel drug delivery systems. Herein, sterically stabilized nanoscale dispersion loaded with curcumin (nanodispersion) based on non-ionic colloidal system has been proposed. In this study, the process conditions were effectively optimized using response surface methodology (RSM) with Box–Behnken design (BBD). The suggested optimum formulation proved to be an excellent fit to the actual experimental output. STEM images illustrate that the optimal curcumin-loaded nanodispersion has spherical morphology with narrow particle size distribution. Particle size distribution study confirms that the solution pH does not affect the nanodispersion, and physical stability study shows that the colloidal system is stable over 90 days of storage at ambient conditions. More importantly, controlled release profile was achieved over 72 h and the *in vitro* drug release data fit well to Higuchi model (*R*^{2}=0.9654).

## Introduction

Nanoencapsulation systems have been widely discussed in the past decade for the delivery of poorly water-soluble compounds especially in the current clinical use of orally administered drugs [1]. In fine particulate system, particle stability is essential as any deviation from optimal particle stability could potentially brings up significant negative impact to its therapeutic performance, i.e., cell uptake and bioavailability could be affected dramatically. Current trends focusing on emulsion-based delivery systems have unlocked the potential in cancer therapy owing to their notable breakthrough in targeted drug delivery and controlled release strategy. However, all emulsion systems are inherently thermodynamically unstable, thus excessive stabilizer is always required to ensure surplus in surface energy to receive stability.

Comparatively, other than colloidal system, solid-based or semi-solid based nanoscale dispersion systems have received less attention in the current clinical research scenario, mainly due to their complexity in preparation, excessive drug payload, and discouraging drug carrier stability [2]. In this study, curcumin, a yellow natural polyphenolic phytochemical extracted from *Curcuma longa* (commonly known as turmeric) which is found to have significant bio-pharmacological activities has been employed as a model lipophilic compound to be incorporated into nanodispersion [3]. However, due to its extensive metabolism in GI tract [4] and its extremely poor solubility in water [5], low oral bioavailability is reported which limits its usage and presents a major challenge in clinical use. Many attempts to resolve these issues through a variety of formulations and elevating it as a possibile next generation anti-cancer therapy were reported. For instance, solubility and bioavailability of curcumin were significantly improved through different formulations such as, curcumin oil-in-water nanoemulsion [6], curcumin-loaded PLGA nanoparticles [7], curcumin solubilised in micelles [8] and loading of curcumin onto modified starch [9]. There is one major setback observed in most of the aforementioned methods which is the involvement of tedious preparation processes to ensure its biological compatibility after administration. Therefore, intensive studies were carried out to overcome the tedious process conditions such as, self-nanoemulsifying drug delivery systems (SNEDDS) which suggests transient formation of drug loaded nanoemulsion system. However, SNEDDS usually demand a relatively higher concentration of surfactant (normally 20–50%), which could possibly raise several potential health issues. Yet, with application of ultrasound these issues could be radically resolved [10, 11].

Development of pharmaceutical formulations is not only time-consuming but also a costly process. Response surface methodology (RSM) is an effective and satisfactory multivariate statistical technique commonly used in process optimization and improvement [12]. Box–Behnken design (BBD) is a reduced three level factorial design proposed by George Box and Donald Behnken in 1960 [13]. This experimental design model has been employed in the development of pharmaceutical formulations such as lacidipine microemulsion [14], cyclosporine self-nanoemulsified drug delivery systems [15], nanosuspension prepared by ball milling [16], dopamine nano-scaffold [17], etc. In the present investigation, a successful attempt has been made to produce curcumin-loaded nanodispersion based on the previously developed curcumin micellar system [18].

## Materials and methods

### Materials

Curcumin in a yellow powder (≥94% of curcuminoid content, ≥80% of curcumin) was obtained from Sigma-Aldrich (M) Sdn Bhd. Sorbitan monolaurate (Span 20, HLB=8.6) and polyoxyethylene (10) cetyl ether [Brij 56, C_{16}H_{33}O(CH_{2}CH_{2}O)_{10}H, HLB=12.9] were purchased from Merck (M) Sdn Bhd. Ethanol (99.5%) was purchased from Fisher Scientifics (M) Sdn Bhd. All the chemicals were used as-received. Distilled water was obtained from Milli-Q^{®} Plus apparatus (Millipore, Billerica, MA, USA).

### Generation of curcumin nanodispersions

Curcumin-loaded nanodisperson was generated using a two-step strategy as shown in Fig. 1. First, curcumin was weighed and dissolved in ethanol to obtain an organic phase at a concentration of 1 mg/g; the mixture was then subjected to vortex mixing for 5 min to ensure complete dissolution. The dissolved curcumin was added to distilled water containing Brij 56 and Span 20 with the predetermined ratio, after that the mixture was vigorously mixed until a transparent isotropic curcumin micellar system was obtained. Ultrasonication (ultrasonic bath, 38 kHz, Ultrasonic RMS 140 W, Guyson International Ltd, UK) was further applied to this curcumin micellar system for 5 min for the size reduction. In the second step, ethanol was removed from the curcumin micelles via diffusion-evaporation using a rotary evaporating flask under reduced pressure (150 mbar) and at a temperature in between 40 and 50°C. Evaporation was varied from 10 to 30 min and the resultant dispersion was collected and immediately diluted with millipore water (10×) followed by a gentle mixing of 10 s. It was then again subjected to sonication using the ultrasonic bath for 5 min to induce fine homogenization. The resultant samples were then immediately analysed using Zetasizer (Malvern, UK), and were sealed within screw-capped bottles and kept under room temperature to study physical stability. All the samples were prepared according to weight measurements with ±0.005 g error.

### Particle size, polydispersity index (PDI) and zeta potential measurements

Dynamic light scattering (DLS) is a common technique for the measurement of suspended solid or liquid particles which measures the intensity of scattering light rebounds from these particles moving under Brownian motion. Mean particle size (or z-average) and polydispersity index (PDI) were comprehensively determined using Stokes-Eintein relationship by employing Zetasizer-Nano ZS (Malvern Instruments Inc., UK) with He-Ne laser 633 nm at a scattering angle of 173°. PDI is usually monitored to have a value not higher than 0.25 to ensure monodipersed population. All the measurements were repeated three times at 25°C.

### Chemical stability study

The stability of nanodispersion at different pH i.e., 2–8 was conducted. The mean particle size was determined by DLS while the surface charge was determined by laser doppler electrophoresis using the Zetasizer Nano ZS (Malvern Instruments Inc., UK). Both measurements were carried out in sequence assisted by an automated titrator (MPT-2).

### Physical stability study

To evaluate the physical stability of curcumin nanodispersion, the clarity of samples and their size distribution were recorded after 10 days of storage for short-term stability study, while the particle size of selected samples was observed up to 90 days for long-term stability study. All the samples were sealed immediately after measurements. Anti-sedimentary and creaming characteristics of curcumin nanodispersions were ensured by subjecting them to 20 min of centrifugation (centrifuge 5810R, Eppendorf AG, Hamburg, Germany) at 5000 rpm and at 25°C. None of them were showing precipitation nor creaming.

### Experimental design

Design Expert^{®} 8.0 (Stat-Ease Inc., Minneapolis, MA, USA) was employed for experimental design, statistical analysis, building the quadratic response surface and for process optimization. Preliminary results [18] revealed that blending of non-ionic surfactants plays a crucial role to control the physical stability of droplets in the colloidal systems, as well as in the determination of size distribution of fine paticles. It was suggested that evaporation temperature and time of application control the rate of solvent removal from the core of curcumin-loaded micelles. To study the synchronized impact of two surfactants in the system, hydrophilic-lipophilic balance (HLB) number of surfactant mixture could be estimated from the following (Eq. 1).

where, N_{HLB,mix} is the HLB number of surfactant mixture, N_{HLB,SurfA} is the HLB number of surfactant A, W_{surf A} is the weight ratio of surfactant A to the total weight of surfactant; while N_{HLB,Surf B} is the HLB number of surfactant B, W_{surf B} is the weight ratio of surfactant B to the total weight of surfactant. In this study, the impact of three independent variables i.e., hydrophilic-lipophilic balance (HLB) number (X_{1}), evaporation temperature (X_{2}) and time of evaporation (X_{3}) were intensively evaluated by Box–Behnken statistical design (BBD). Through this design, 17 experiments were suggested consisting of five replicated center points and a set of scattering points lying at the center of each edge of the 3D cube, as shown in Fig. 2. It is essential to repeat the center point for five times to ensure that the reproducibility of dispersion was achieved. Multiple regression analysis to fit the second order polynomial equaition (Eq. 2) was statistically evaluated and validated using analysis of variance (ANOVA). Four dependent responses were investigated, namely particle size (Y_{1}), polydispersity index (Y_{2}), zeta-potential (Y_{3}) and zeta-average after 10 days of storage (Y_{4}), and the obtained responses could be described in 2D surface by using a generalised response function as below (Eq. 2).

where Y is the respective response; β_{0} is a constant; β_{i}, β_{ij}, β_{ii} are linear, interaction and quadratic coefficients, respectively. To obtain the best fitting quadratic mathematical model, optimization was further continued to obtain the nanodispersion with the smallest possible particle size, lowest polydispersity index, largest negative zeta-potential and the lowest zeta-average after 10 days of storage.

### Scanning transmission electron microscopy (STEM) studies

The morphology and the particle size of nanodispersions were examined using field emission scanning electron microscopy (FE-SEM, FEI Quanta 400F ESEM with EDAX, USA) in STEM mode. A drop of freshly prepared curcumin-loaded nanodispersion was placed on a 50-mesh copper grid and the excess liquid was removed using Whatman filter paper. The samples were then allowed to dry under room temperature for 5 min and then a drop of phosphotungstic acid (PTA, 3%) was added and kept for 5 min for staining prior to STEM investigation.

*In vitro* dissolution test

Controlled drug release pattern of curcumin in simulated gastric fluid (pH 1.2) was studied using the USP XVIII method with USP 1 rotating paddle apparatus. Briefly, 5 mL of curcumin-loaded nanodispersion was filled in a dialysis bag (Sigma, with the molecular weight cut-off of 12 000 Da), tightly closed and placed at the bottom of a dissolution flask containing the dissolution medium with a paddle rotating at 100 rpm. The temperature of dissolution medium was controlled at 37°C by immersing the vessel in a water bath equipped with an external temperature control unit. The amount of curcumin released to the dissolution medium was analysed using UV-Vis spectrophotometer (PerkinElmer Lambda 35), as earlier reported [19–21]. One milliliter of sample was withdrawn from the dissolution medium at regular intervals and analysed spectrophotometrically at 425 nm, and an equivalent volume of buffer (pH 1.2) was replaced. The dissolution test was performed in triplicate. The cumulative release of curcumin (percentage) released at different time intervals was calculated using the following (Eq. 3).

## Results and discussion

### Statistical study of regression model

Box–Behnken design (BBD) is a reduced statistical experimental design which selects relevant points from a three level factorial arrangement. The experimental points are spread spherically equidistant from the center point and the experimental number (*N*, or the number of experiments) needed to build-up the regression model can be determined by Eq. 4.

where *k* is the number of independent variables (or factors), and *Cp* is the number of experiments for center point. In comparison to an ordinary three level factorial design suggesting *N*=*k*^{3}, the total number of experiments is 27 which is 58.82% more than the Box–Behnken design. The total number of experiments needed for the Box–Behnken design is only 17 by assuming the number of experiment for center point is 5.

From the preliminary experiments, HLB number, evaporation temperature and time of evaporation were realized to be the most significant variables to determine the characteristics of curcumin-loaded nanodispersion, i.e., HLB number suggests the curvature of liquid–liquid interface which indirectly determines the stability of particle formation, whereas the evaporation temperature and duration control the formation of particles. Three design variables were controlled at three levels (–1, 0, +1) corresponding to the experimental domain which determine the minimum and maximum coded value of experimental variables studied and the actual value of experimental variables that can be determined according to the following Eq. 5.

where *x*_{i} is the coded value of experimental variable, *z*_{i} is the actual value of experimental variable, *z*_{0} is the center point of experimental variable and Δ*z* is the interval of variation. The response surface model was determined after fitting the response data into the function and the validity was statistically evaluated with analysis of variance (ANOVA). Out of the several mathematical models i.e., linear, two-factor interaction, quadratic and cubic which attempt to fit the experimental data, quadratic model is the most reliable model to describe the process due to a smaller given predicted residual sum of squares (PRESS) as suggested by the simulation. Four responses were measured and the residual (RES, which is the difference between the actual experimental value and the predicted value) has been shown in Table 1.

Run order | Particle size | Polydispersity index | Zeta-potential | Z-average after 10 days | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|

Actual value | Predicted value | Residual | Actual value | Predicted value | Residual | Actual value | Predicted value | Residual | Actual value | Predicted value | Residual | |

1 | 233.7 | 289.488 | –55.788 | 0.459 | 0.44113 | 0.01788 | –36.1 | –34.988 | –1.1125 | 116.4 | 89.9988 | 26.4013 |

2 | 587.7 | 531.913 | 55.7875 | 0.688 | 0.70588 | –0.0179 | –10.1 | –11.213 | 1.1125 | 411.6 | 438.001 | –26.401 |

3 | 207.3 | 208.325 | –1.025 | 0.442 | 0.41363 | 0.02838 | –21 | –22.425 | 1.425 | 95.69 | 120.33 | –24.64 |

4 | 250.1 | 189.64 | 60.46 | 0.404 | 0.329 | 0.075 | –23.5 | –23.02 | –0.48 | 119.2 | 156.5 | –37.3 |

5 | 329.8 | 353.763 | –23.963 | 0.469 | 0.39213 | 0.07688 | –36.2 | –35.713 | –0.4875 | 127.1 | 78.3263 | 48.7738 |

6 | 138.5 | 171.35 | –32.85 | 0.274 | 0.30463 | –0.0306 | –22.4 | –23.2 | 0.8 | 100.4 | 147.413 | –47.013 |

7 | 169.8 | 189.64 | –19.84 | 0.423 | 0.329 | 0.094 | –24.9 | –23.02 | –1.88 | 125.3 | 156.5 | –31.2 |

8 | 168.3 | 167.275 | 1.025 | 0.448 | 0.47638 | –0.0284 | –22.8 | –21.375 | –1.425 | 279.4 | 254.76 | 24.64 |

9 | 574.4 | 550.438 | 23.9625 | 0.653 | 0.72988 | –0.0769 | –12.2 | –12.688 | 0.4875 | 159.4 | 208.174 | –48.774 |

10 | 117.4 | 84.55 | 32.85 | 0.471 | 0.44038 | 0.03063 | –23.7 | –22.9 | –0.8 | 110.7 | 63.6875 | 47.0125 |

11 | 192.8 | 189.64 | 3.16 | 0.294 | 0.329 | –0.035 | –22 | –23.02 | 1.02 | 180.5 | 156.5 | 24 |

12 | 189.6 | 132.788 | 56.8125 | 0.421 | 0.46725 | –0.0463 | –37.5 | –37.188 | –0.3125 | 106.5 | 108.261 | –1.7613 |

13 | 164.5 | 189.64 | –25.14 | 0.254 | 0.329 | –0.075 | –22.4 | –23.02 | 0.62 | 148.2 | 156.5 | –8.3 |

14 | 231.6 | 288.413 | –56.813 | 0.914 | 0.86775 | 0.04625 | –12.8 | –13.113 | 0.3125 | 374.3 | 372.539 | 1.76125 |

15 | 479.8 | 502.738 | –22.938 | 0.73 | 0.6815 | 0.0485 | –15.2 | –13.288 | –1.9125 | 482.8 | 409.386 | 73.4138 |

16 | 171 | 189.64 | –18.64 | 0.27 | 0.329 | –0.059 | –22.3 | –23.02 | 0.72 | 209.3 | 156.5 | 52.8 |

17 | 242.2 | 219.263 | 22.9375 | 0.431 | 0.4795 | –0.0485 | –34.1 | –36.013 | 1.9125 | 122.4 | 195.814 | –73.414 |

The degree of freedom for the square of regression in the current study is 9 and the degree of freedom of each variation source is determined to be 9, 7, 3, 4, 16 for regression model, residuals, lack of fit, pure error and total, respectively. The significance of regression can be determined using the Fisher distribution (F test), as shown in Eq. 6,

where MS_{reg} is the media of the square of regression and MS_{res} is the media of the square of residuals. By referring to all the *F*-values collected from ANOVA (Table 2), the mathematical model for all the four responses are shown to be significant. For instance, the response for particle size (Y_{1}) having a model regression *F*-value of 11.28 implies that only a 0.21% opportunity for the failure to occur due to noise. The validity of the model can also be evaluated using lack of fit test, as suggested by the following Eq. 7,

Particle size (nm) | Polydispersity index (PDI) | Zeta-potential | Z-average after 10 days | |||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|

Regression coefficient | F-value | p-Value | Regression coefficient | F-value | p-Value | Regression coefficient | F-value | p-Value | Regression coefficient | F-value | p-Value | |

R^{2} | 0.9355 | 0.9032 | 0.9810 | 0.8783 | ||||||||

Model | 11.277 | 0.0021 | 7.2551 | 0.008 | 40.221 | <0.0001 | 5.6151 | 0.0166 | ||||

Lack of fit | 4.5095 | 0.0899 | 1.3406 | 0.3795 | 3.6913 | 0.1197 | 5.1912 | 0.0727 | ||||

α0 | 189.64 | 0.329 | –23.02 | 156.5 | ||||||||

Linear | ||||||||||||

α1 | 109.78 | 30.564 | 0.0009 | 0.1506 | 25.554 | 0.0015 | 11.7 | 356.38 | <0.0001 | 119.46 | 28.171 | 0.0011 |

α2 | 31.963 | 2.5911 | 0.1515 | –0.05 | 2.7737 | 0.1398 | –0.338 | 0.2965 | 0.603 | –12.68 | 0.3172 | 0.5909 |

α3 | 11.438 | 0.3318 | 0.5826 | –0.018 | 0.3751 | 0.5596 | 0.1875 | 0.0915 | 0.771 | 54.539 | 5.8715 | 0.0459 |

Interaction | ||||||||||||

α12 | 75.2 | 7.1715 | 0.0316 | –0.044 | 1.0656 | 0.3363 | 0.25 | 0.0814 | 0.7837 | 31.1 | 0.9546 | 0.3611 |

α13 | –20.7 | 0.5434 | 0.485 | 0.0063 | 0.022 | 0.8863 | 0.55 | 0.3938 | 0.5502 | 60.375 | 3.5977 | 0.0997 |

α23 | –29.93 | 1.1356 | 0.3219 | –0.036 | 0.74 | 0.4181 | –0.575 | 0.4304 | 0.5328 | –41 | 1.6589 | 0.2387 |

Quadratic | ||||||||||||

α11 | 184.84 | 45.609 | 0.0003 | 0.2268 | 30.479 | 0.0009 | –1.528 | 3.197 | 0.1169 | 86.039 | 7.6909 | 0.0276 |

α22 | –88.68 | 10.498 | 0.0143 | 0.0683 | 2.7613 | 0.1405 | –0.353 | 0.1703 | 0.6922 | 28.961 | 0.8714 | 0.3816 |

α33 | 56.918 | 4.3245 | 0.0761 | 0.0115 | 0.0784 | 0.7876 | 0.8975 | 1.1037 | 0.3284 | –38.91 | 1.5732 | 0.25 |

where MS_{lof} is the media of the square of lack of fit and MS_{pe} is the media of the square of pure error. For the response of particle size (Y_{1}), lack of fit *F*-value was calculated to be 4.5095, indicating a 8.99% probability that a “lack of fit” could occur, slightly higher than the recommended value which is commonly to be 5%.

The most common approach to validate a suggested model is through the linear correlation plots obtained using the predicted and actual experimental data. In Fig. 3, a higher *R*^{2} was obtained, i.e., 0.9355, 0.9032, 0.9810 and 0.8783 for particle size, polydispersity index, zeta potential and zeta-average after 10 days of storage, respectively. Although, Fig. 3d shows a comparatively low *R*^{2}, the suggested model is still applicable to explain the dependence of particle growth over the process variables. Generally, the collected data from designed experiments are close to the predicted values which demonstrate the suitability of the mathematical model. To further investigate the functional relationship for approximation, the second order polynomial regression equations for each response are suggested as shown in the following Eqs. 8–11.

To determine the influence of each independent variable to the dependent variables, *p*-value of each independent variable as well as the interaction between independent variables and their quadratic values have been shown in Table 2. Due to its ability to control interface curvature, HLB number has a *p*-value <0.05 in all the responses and found to be the most critical parameter to determine the final outcome of curcumin-loaded nanodispersion. With combined use of two (or more) different surfactants, it gives rise to a better drug loading for hydrophobic compounds in water [22]. The optimal HLB number to form oil-in-water emulsion was observed to fall within the range from 10 to 12 [23, 24]. Although Griffin’s assumption may not strictly reflect the true situation, it is considered as a good approximation in most of the studies. RSM offers an opportunity to visualize the interaction among several process parameters on the responses. From Fig. 4, it could be seen that an increase in HLB number could significantly lead to an increase in all the responses. However, the surface model reveals a non-linear behavior between HLB number (X_{1}) and evaporation temperature (X_{2}) for Y_{1}, suggesting a typical saddle surface which confines the desired process conditions within a smaller zone, as indicated in the region with navy color. This is basically in line with Orafidiya and Oladimeji [25] for the similar observation. From Fig. 4, it could also be observed that a higher HLB number leads to the formation of nanodispersions with larger size which could be attributed to the presence of a long chain hydrocarbon in Brij 56 as compared to Span 20. Generally, solvent evaporated rapidly within the first few minutes followed by a slow and gentle evaporation until its complete removal, basically explaining why time of evaporation (X_{3}) is not significant in all the responses except for Y_{4}. The quadratic model evidently reveals that the evaporation temperature in the range of 40–50°C is adequate for efficient removal of solvent without any degradation of the encapsulated curcumin. Nevertheless, a higher evaporation temperature above 50°C which led to an aggresive solvent removal may result in particle destruction. As evaporation duration is not a constraint in this study, evaporation was allowed up to 30 min to ensure complete solvent removal.

The experimentally obtained zeta-potential (Y_{4}) was approximated by a quadratic equation. Extremely low *p*-value of <0.0001 indicates that the model is perfectly fitted with the actual experimental data (Table 2). A linear relationship between HLB number and evaporation temperature, as well as HLB number and evaporation time was observed (Fig. 4). It is evident that the static charges on the particle surface is controlled by the HLB number of non-ionic surfactant (*F*-value 356.38, *p*-value <0.001). It is generally agreed that the lower polydispersity index and the lower zeta-potential have a direct relationship to slow down the particle growth or to lower the rate of Ostwald ripening [26]. As shown in Table 2, the quadratic function describing the zeta-average of samples after 10 days of storage has a *p*-value as low as 0.0166, indicating that the data is reliable to represent the actual experiment. The mathematical model clearly shows that the particle growth is affected by an initial particle size distribution; a lower polydispersity index suggests a narrow particle size distribution, could vitally moderate the dynamics of Ostwald ripening governed by Gibbs–Thomson effect. This phenomenon explains the tendency to dissolve smaller particles and the transfer of mass to the larger particles [27]. A further physical stability study based on Y_{4} was carried out and discussed in the following Section 3.4.

### Optimisation of curcumin-loaded nanodispersions

The optimization was perormed based on the criteria as stated in Table 3. Considering several process variables to fulfill the desired criteria and process feasibility, a given optimum formulation with the desirability of 0.9172 was employed in this study. It is suggested that, 1% surfactant with the HLB number of 9.89 (Brij 56 and span 20 in a weight ratio of 3:7), 27 min of solvent evaporation at 50°C and under 150 mbar lead to the formation of curcumin nanodispersion with the desired particle size, polydispersity index, zeta-potential and zeta-average after 10 days of storage to be 117.39 nm, 0.341, –32.44 mV and 71.77 nm, respectively. The adequacy of this simulated model for optimum response prediction has been confirmed by the experiments. The accuracy of predicted formulation could be described by the following Eq. 12.

Goal | Suggested | Experimental | Accuracy (%) | |
---|---|---|---|---|

X_{1}: Hydrophilic-lipophilic balance (HLB) number | – | 9.89 | 9.89 | – |

X_{2}: Temperature | – | 50 | 50 | – |

X_{3}: Duration | – | 27.13 | 27 | – |

Y_{1}: Particle size | Minimized | 117.39 | 108.50 | 92.41 |

Y_{2}: Polydispersity index (PDI) | Minimized | 0.341 | 0.398 | 83.30 |

Y_{3}: Zeta-potential | Minimized | –32.44 | –30.50 | 94.01 |

Y_{4}: Z-Average after 10 days | Minimized | 71.77 | 73.69 | 97.33 |

A good agreement could be noted as shown in Table 3. The suggested formulation shows high accuracy for each of the responses, Y_{1}, Y_{2}, Y_{3} and Y_{4}, i.e., 92.41%, 83.30%, 94.01% and 97.33%, respectively. Response surfaces and contour plots for the optimization function describe a multiple optimization strategy and determine the level of each independent variable to reach the highest desirability. As indicated from Table 3, an optimum formulation with the highest desirability could be obtained if (1) the HLB number of surfactant is fixed at a value of approximately 10, (2) the applied evaporation temperature is as high as possible without destroying the particles, and (3) preferably a longer time of evaporation is employed. While time of evaporation is found to be an insignificant variable in this study, the process can be terminated when no bubbles are observed from the surface during evaporation. However, a longer evaporation is always preferred, or more specifically 20 min after the last bubble captured with naked eyes, is always preferred to ensure complete removal of solvent, provided that the energy consumption and time are not the constraints. Field emission scanning electronic microscopy (FESEM) was employed to study the surface morphology of curcumin nanodispersion obtained under the best optimum conditions. As illustrated in Fig. 5, the curcumin-loaded nanodispersion consists of two parts; curcumin rich zone which is the dark region in the core, surrounded by a layer of surfactant. Particle size distribution obtained from this study is in close agreement with the particle size measurements from Malvern Zetasizer-Nano.

### Chemical stability study

Chemical stability study for the obtained optimal curcumin-loaded nanodispersion at different pH from 2 to 8.5 was carried out to evaluate the effect of solution pH on the particle surface charge (zeta-potential), which indirectly governs the z-average of particles and hence the physical stability of the system (Fig. 6a). Polydispersity index represents the size distribution of nanodispersions at different pH as shown in Fig. 6b. This study is important as the formulated nano-scale drug carriers are expected to expose to various compartments of body, as well as to resist a wide range of pH throughout the gastrointestinal tract (GI tract) before it releases curcumin to the targeted area. Similar findings have been reported, e.g., Dixit and Nagarsenker [28] achieved the smallest possible nano globule in a buffer with the pH of 2.5, whereas the globule size was increased when a buffer with pH 6.8 was utilized. This is probably due to higher concentration of hydrogen ions in the buffer with the pH 2.5, suggesting a larger electrostastic repulsion subjected to globule surface, pushing globule to shrink which resulted in the smallest possible size. In this study the obtained z-average is approximately 125 nm and an insignificant decrease in terms of z-average was observed with an increase in pH. This is due to that an increase in the pH leads to a higher tendency to attract more hydrophobic residues onto the particles surface, indirectly improving the emulsifying capacity, and thus resulting into a notable decrease in the interfacial tension. This occurs when the particles are in a relatively smaller size. However, a reduction in the particle size is considered to be negligible in this investigation. For z-average, a relatively larger fluctuation was observed in the lower pH regime and a similar phenomenon was observed in the measurement of polydispersity index (Fig. 6b). Despite a fluctuation in the polydispersity index of nanodispersion within the designated range of pH, it is still within in an acceptable narrow range i.e., 0.23–0.39. Nanodispersions can be said to be negatively charged in which the charge is substantially contributed by the pH of buffer. It was found that an increase in the pH resulted in a gradual decrease in the zeta-potential, and a drastic reduction was observed after pH 5. Liu et al. [24] have reported similar observations. Excellent narrow particle size distribution together with a large negative zeta-potential are crucial to lower the rate of the Ostwald ripening effect. Under the conditions of room temperature, the particle growth rate within 10% was observed after 10 days of storage. This shows the rate of flocculation is controllable with this proposed nanodispersion system.

### Physical stability study

To date, very limited research studies have addressed the physical stability issue of drug-loaded nanodispersion, particularly on the change in size over a long-term storage. Physical stability of particles can be referred to any change in the measureable attributes, but generally refers to the ability to keep suspended particles or droplets in a liquid medium without severe sedimentation, creaming, flocculation and coalescence. Thus, this work attempts to explain the observed phenomenon for 90 days of physical stability.

Suspended solid particles tend to conglomerate themselves by chemical bridging, e.g., van der Waals force to create minimum surface area to release surface energy. To enhance the stability of system, chemical bridging can be reduced via adsorption of ionic surfactants onto particle surface to create mutual repulsion, nevertheless the use of ionic surfactants are always harmful to human body and thus their incorporation is not always encouraged. Alternatively, the dispersion stability against aggregation/flocculation/coalescence can be achieved by a thin layer of non-ionic dynamic coating to acquire steric stabilization. Though non-ionic surfactants can hardly produce a stable electrostatic protection on the surface of dispersed phase, steric stabilization induced by bending of non-ionc surfactants on dispersed phase masks the influence of surface attractive forces. For instance, limiting the impact of van der Waals interaction force between water molecules and dispersed phase can alter the magnitude of attraction and packing structure of the particles, causing them to behave counter intuitively with ordinary solid dispersion system. This is generally regarded as a sterically stabilized colloidal system.

Curcumin-loaded nanodispersion formulation based on RSM experimental design has been extended; process parameters were properly controlled and the reduction in the size of nanodispersion has been presented in Fig. 7. The sterically stabilized curcumin-loaded nanodispersion demonstrates the ability to prevent the growth of particles, though a progressive decrease could be observed with an increase in storage time as a result of drug release. From Fig. 7a, curcumin-loaded nanodispersion with higher curcumin content (α) or generated with higher evaporation temperature is generally holding a relatively larger z-average at the beginning. When the drug release took place, reduction in the size was observed and the release behavior can be described with an exponential equation as shown in Fig. 7a. Similar phenomenon could be observed from Fig. 7b and c. The addition of stabilizer is necessary. However, the amount of stabilizer involved in the formulation can be minimized if the activation energy that triggers all the unfavorable events at the very beginning can be suppressed by the presence of a smaller amount of stabilizer in the system.

Fundamentally, ionized surfactants are always employed to stabilize the dispersed system due to their intrinsic ability to create mutual repulsion among the particles by the deposition of opposite electrical charge on the surface of particles. However, this colloidal stability reinforced by electrostatic stabilization is also subjected to extrinsic forces acting to the system, thus, the stability of this colloidal system can be greatly affected by solution pH which stands a high chance to collapse at very acidic or very alkali condition. Povey and Ding [29] suggested that particles mobility driven by Brownian motion would become more significant if the particles fall into nano scale. This gives rise to an opportunity for smaller particles to counter van der Waals forces which is relatively weak compared to the momentum devoted in particle motion. This further explains why nanoemulsion is said to be kinetically stable. For ordinary dispersed systems stabilized by ionic surfactants, surface charge leads to the formation of double electrical layers which do not induce steric barrier to regulate the mass transfer. The double electrical layers at the surface boundary can be analyzed and the study of this electrical potential is generally referred as zeta potential. It is suggested that higher the negative zeta potential greater the repulsion force among the particles, and hence a better colloidal stability. Capek [30] suggested that the thickness of non-ionic surfactant on the interface also contributes to zeta potential. This is in line with the current finding as stated in the chemical stability study.

In this study, curcumin-loaded nanodispersion has been generated using a mixture of two non-ionic surfactants, and the stability of colloidal system is primarily governed by steric stabilization. This is particularly of great interest in pharmaceutical applications since most of the ionic stabilizers are toxic to human body. In short, the currently employed sterically stabilized dispersed system can be described in two parts [31, 32]. First, osmotic interaction induced by thickness of non-ionic adsorbed layer and volume fraction of polymer chain, suggesting a steric repulsion as described in the following Eq. 13.

where *G*_{mix} denotes the osmotic interaction, *k* is the Boltzmann constant, *T* is the absolute temperature, *V*_{1} is the molar volume of the solvent, *φ*_{2} is the volume fraction of polymer chain in the adsorbed layer, *X* is the Flory-Huggins parameter (polymer-solvent interaction), *δ* is the thickness of adsorbed layer and *h* is the distance of separation between the particles. Second, the reduction in the configurational entropy of polymeric chains can be explained in terms of elastic interaction as shown in the following Eq. 14.

where *G*_{el} is the elastic interaction, *v*_{2} is the number of polymeric chains per unit area, Ω(*h*) is the configurational entropy of polymeric chain at distance *h*, and Ω(∞) is the configurational entropy of the polymeric chain at an infinite distance. The two equations explain the scenario of repulsion that occurs between two sterically stabilized particles. When two particles collide, non-ionic surfactants on the particle surface condense which results in an increase in the particle interface configurational entropy; the above builds up the elastic potential and thus repels each other until the original interfacial configurational entropy is reached.

Among all the unfavored factors that cause the colloidal system physically unstable, irreversible flocculation is said to be the most severe phenomena which should be avoided in the current solid nanodispersion system. Smoluchowski [33] attempted to describe the phenomena which was later strengthened [34] as given in the Eq. 15.

where *n*_{a} is the total number of aggregates per unit volume, *n*_{0} is the initial number of aggregates, *k*_{1} and *k*_{2} are the factors which describe the average stability ratio of particles. In comparison to several exponential equations obtained after 90 days of storage which tentatively describe the reduction in the z-average, it is evident that severe flocculation was not occurred within 90 days. This might be attributed to redistribution of surfactant molecules on the nano particle surface as a result of a progressive decrease in the total surface area by assuming that the total number of nanoparticles in the liquid medium does not change with time.

Ostwald ripening is another destabilization that occurs especially in the solid dispersed systems. It occurs only when the solid dispersed phase holds a relatively higher solubility in the liquid medium [35]. Larger particles consume smaller particles to form very large particles. This peculiar phenomenon can be explained with the LSW theory [36, 37] by using the following Eq. 16.

where 〈*R*〉 is the average radius of particles in the system, *γ* is the surface tension of particle, *c*_{∞} is the solubility of the particle, *v* is the molar volume of the particle, *D* is the diffusive coefficient of particle and *R*_{gas} is the ideal gas constant. Ostwald ripening is generally found in the dispersed system with low negative zeta potential. However, no evidence indicates that Ostwald ripening was taking place during the 90 days of storage, as from the Eq. (16) z-average should increase with time but in the current investigation particle size was found to decrease with an increase in time.

*In vitro* release study

Curcumin did not show any release [38] or a burst release within a very short time [39]. These studies were conducted using simulated body fluid. *In vitro* drug release of curcumin-loaded nanodispersion from RSM was investigated. Several release models were employed to describe the release kinetics and the obtained results have been shown in Table 4. By comparing the numerical significance, Higuchi (Fig. 8) and Korsmeyer–Peppas models have been proposed to have the best fit among all, which hold a siginicantly higher *R*^{2} i.e., 0.9654 and 0.9643, respectively. In short, the release mechanism is concentration-dependent as a relatively lower *R*^{2} was observed for the zero order release model; and the drug released via diffusion due to a relatively higher *R*^{2} was observed from Higuchi model. The controlled release strategy proposed by a thin steric barrier composed by a mixture of two non-ionic surfactants seems to be quite promising. This not only encourages a steady release of hydrophobic curcumin in the body fluid but also restrains a burst release immediately after administration. The Korsmeyer–Peppas model further suggests that the kinetics of release is governed by Fickian diffusion (*n*=0.0818). Previous studies indicate that the encapsulated drug in the solid particle could promote an enhanced drug immobilization and a slower release rate [47, 48]. This has been attributed to their drug-polymer complex induced exclusion from the dissolution medium. It also acts as a physical barrier to protect the encapsulated drug from degradation. Setthacheewakul et al. [49] developed curcumin self-microemulsifying drug delivery system (SMEDDS) which provided the best release profile when the drug formulation was introduced to simulated gastric fluid of pH 1.2. Disintegration of curcumin SMEDDS pellet took place within 5 min, and an immediate release from the pellet which formed a fine O/W microemulsion with a transparent appearance showed a constant release rate of up to 120 min. Higher surface area of the drug loaded nanodispersion was found to demonstrate an enhanced release profile. However, a rapid complete release could potentially give rise to few drawbacks, e.g., a frequent dose is needed due to the difficulty in maintaining the therapeutic effect to an extended period. Moreover, a sudden burst of drug release into the blood circulation could potentially result in drug poisoning. In addition, the optimized nanodispersion developed in our studies clearly confirms that the sustained release profile is preferred over other drug carrier systems in terms of drug release control, particularly for a system with the surfactant concentration as low as 1 wt/wt%.

Release model | Description | Release constant | R^{2} | Ref. |
---|---|---|---|---|

Zero order | where Q_{t} is the amount of drug dissolved in time t, K_{0} is the zero order release constant | K_{0}=0.2933 | 0.8731 | [40] |

First order | or K_{1} is the first order release constant | K_{1}=–0.005987 | 0.9038 | [41] |

Hixson-Crowell | K_{s} is the Hixson–Crowell release constant incorporating the surface–volume relation | K_{S} =0.0073 | 0.8940 | [42] |

Higuchi | K_{H} is the Higuchi release constant, which describes the release as a diffusion process governed by Fick’s Law | K_{H} =2.7182 | 0.9654 | [19, 43, 44] |

Korsmeyer-Peppas | K_{K} is the Korsmeyer–Peppas release constant. This model sometimes referred as Power Law model | K_{K} =42.01458n=0.0818 | 0.9643 | [45, 46] |

## Conclusion

The fabrication of curcumin-loaded nanodispersion has been successfully accomplished with response surface methodology. The obtained quadratic mathematical model from the Box–Behnken design (BBD) is in good agreement with the experimental data, besides showing a good fit between the predicted and the actual optimized formulation. The present study demonstrates that the curcumin-loaded nanodispersion can be conveniently prepared from curcumin micelles by monitoring the colloidal system’s HLB number. Particle size distribution and zeta-potential are the crucial parameters that ensure superior stability of the formulation over long-time storage. A narrow particle size distribution could be effectively achieved within few minutes using intensive ultrasonication, and the largest possible negative zeta-potential could be obtained by simply controlling the combination of two surfactants in the formulation. A higher evaporation temperature under the reduced pressure remains the critical factor to accelerate a rapid evaporation of solvent, yet the shortest time of evaporation is suggested as it ensures cost-efficiency without consuming a larger amount of energy. Based on 90 days of storage and dissolution profile, it suggests that curcumin-loaded nanodispersion could be a good candidature for the next generation drug formulation.

## Article note:

A collection of invited papers based on presentations at the 5^{th} international IUPAC Conference on Green Chemistry (ICGC-5), Durban (South Africa), 17–21 August 2014.

## Acknowledgments

Authors would like to thank Fundamental Research Grant Scheme (FRGS) for the funding support (Grant/Award Number: FRGS/1/2013/SG05/UNIM/01/1).

## References

[1] C. Pinto Reis, R. J. Neufeld, A. J. Ribeiro, F. Veiga. *Nanomed: Nanotech. Bio. Med.***2**, 53 (2006).Search in Google Scholar

[2] S. Verma, S. Kumar, R. Gokhale, D. Burgess. *Int. J. Pharm.***406**, 145 (2011).Search in Google Scholar

[3] C. P. Tan, M. Nakajima. *Food Chem.***92**, 661 (2005).Search in Google Scholar

[4] C. R. Ireson, D. J. Jones, S. Orr, M. W. Coughtrie, D. J. Boocock, M. L. Willian, P. B. Farmer, W. P. Steward, A. J. Gescher. *Cancer Epidem.**Biomar. Prev.***11**, 105 (2002).Search in Google Scholar

[5] H. H. Tønnesen, M. Másson, T. Loftsson. *Int. J. Pharm.***244**, 127 (2002).Search in Google Scholar

[6] W. Wang, Y. Jiang, Y. W. Wang, M. T. Huang, C. T. Ho, Q. Huang. *Food Chem.***108**, 419 (2008).Search in Google Scholar

[7] P. Anand, H. B. Nair, B. Sung, A. B. Kunnumakkara, V. R. Yadav, R. R. Tekmal, B. B. Aggarwal. *Biochem. Pharmacol.***79**, 330 (2010).Search in Google Scholar

[8] D. Suresh, K. Srinivasan. *Food Chem. Toxicol.***45**, 1437 (2007).Search in Google Scholar

[9] H. Yu, Q. Huang. *Food Chem.***119**, 669 (2010).Search in Google Scholar

[10] P. R. Gogate, A. M. Kabadi. *Biochem. Eng. J.***44**, 60 (2009).Search in Google Scholar

[11] S. Kentish, T. J. Wooster, M. Ashokkumar, S. Balachandran, R. Mawson, L. Simons. *Innov. Food Sci. Emerg.***9**, 170 (2007).Search in Google Scholar

[12] M. A. Bezerra, R. E. Santelli, E. P. Oliveira, L. S. Villar, L. A. Escaleira. *Talanta***76**, 965 (2008).10.1016/j.talanta.2008.05.019Search in Google Scholar

[13] B. George, B. Donald. *Technometrics***2**, 455 (1960).Search in Google Scholar

[14] R. Gannu, C. R. Palem, V. V. Yamsani, S. K. Yamsani, M. K. Yamsani. *Int. J. Pharm.***388**, 231 (2010).Search in Google Scholar

[15] A. S. Zidan, O. A. Sammour, M. A. Hammad, N. A. Megrab, M. J. Habib, M. A. Khan. *Int. J. Pharm.***332**, 55 (2006).Search in Google Scholar

[16] D. S. Singare, S. Marella, K. Gowthamrajan, G. T. Kulkarni, R. Vooturi, P. S. Rao. *Int. J. Pharm.***402**, 213 (2010).Search in Google Scholar

[17] S. Pillay, V. Pillay, Y. E. Choonara, D. Naidoo, R. A. Khan, L. C. du Toit, V. M. Ndesendo, G. Modi, M. P. Danckwerts, S. E. lyuke. *Int. J. Pharm.***382**, 277 (2009).Search in Google Scholar

[18] K. W. Tan, M. Sivakumar. *Asia Pac. J. Chem. Eng.***7**, 125 (2012).Search in Google Scholar

[19] U. S. Toti, T. M. Aminabhavi. *J. Controlled Release***95**, 567 (2004).10.1016/j.jconrel.2003.12.019Search in Google Scholar

[20] N. Tiong, A. A. Elkordy. *Eur. J. Pharm. Biopharm.***73**, 373 (2009).Search in Google Scholar

[21] B. Boruah, P. M. Saikia, R. K. Dutta. *J. Photochem. Photobiol., A: Chem.***245**, 18 (2012).Search in Google Scholar

[22] S. I. Ahmad, K. Shinoda, S. Friberg. *J. Colloid Interface Sci.***47**, 32 (1974).Search in Google Scholar

[23] W. C. Griffin. *J. Cosmet. Sci.***1**, 311 (1949).Search in Google Scholar

[24] W. Liu, D. Sun, C. Li, Q. Liu, J. Xu. *J. Colloid Interface Sci.***303**, 557 (2006).Search in Google Scholar

[25] L. O. Orafidiya, F. A. Oladimeji. *Int. J. Pharm.***237**, 241 (2002).Search in Google Scholar

[26] J. I. Acedo-Carrillo, A. Rossas-Durazo, Herrera-Urbina, M. Rinaudo, F. M. Goycoolea, M. A. Valdez. *Carbohydr. Polym.***65**, 327 (2006).Search in Google Scholar

[27] G. Madras, B. J. McCoy. *Chem. Eng. Sci.***18**, 3809 (2002).Search in Google Scholar

[28] R. P. Dixit, M. S. Nagarsenker. *Eur. J. Pharm. Sci.***35**, 183 (2008).Search in Google Scholar

[29] M. Povey, Y. L. Ding. Nanodispersion and method of formation thereof. Patent WO 2010038087 A2, Filed 5 Oct 2009, Published 8 April 2010.Search in Google Scholar

[30] I. Capek. *Adv. Colloid Interface Sci.***107**, 125 (2004).10.1016/S0001-8686(03)00115-5Search in Google Scholar

[31] D. H. Napper. *Polymeric Stabilization of Colloidal Dispersions*, Academic Press, London (1983).Search in Google Scholar

[32] T. F. Tadros (Ed.). “Polymer adsorption and colloid stability”, in *The Effect of Polymers on Dispersion Properties,* Academic Press, London (1982).Search in Google Scholar

[33] M. V. Smoluchowski. *Phys. Chem.***92**, 129 (1917).Search in Google Scholar

[34] G. Urbina-Villalba, J. Toro-Mendoza, M. García-Sucre. *Langmuir***21**, 1719 (2005).10.1021/la048433tSearch in Google Scholar
PubMed

[35] T. G. Mason, J. Bibette, D. A. Weitz. *Phys. Rev. Lett.***75**, 2051 (1995).Search in Google Scholar

[36] I. M. Lifshitz, V. V. Slesov. *J. Phys. Chem. Solids***19**, 35 (1961).Search in Google Scholar

[37] Z. Wagner. *Elektrochem.***65**, 581 (1961).10.1001/archopht.1961.01840020583023Search in Google Scholar

[38] J. Cui, B. Yu, Y. Zhao, W. Zhu, H. Li, H. Lou, G. Zhai. *Int. J. Pharm.***371**, 148 (2009).Search in Google Scholar

[39] W. Wu, J. Shen, P. Banerjee, S. Zhou. *Biomaterials***32**, 598 (2011).10.1016/j.biomaterials.2010.08.112Search in Google Scholar
PubMed

[40] P. Costa, J. M. Sousa Lobo. *Eur. J. Pharm. Sci.***13**, 123 (2001).Search in Google Scholar

[41] Y. Hattori, Y. Haruna, M. Otsuka. *Colloids Surf., B***102**, 227 (2013).10.1016/j.colsurfb.2012.08.017Search in Google Scholar
PubMed

[42] N. Tanaka, K. Imai, K. Okimoto, S. Ueda, Y. Tokunaga, A. Ohike, R. Ibuki, K. Higaki, T. Kimura. *J. Controlled Release***108**, 386 (2005).10.1016/j.jconrel.2005.08.024Search in Google Scholar
PubMed

[43] J. Siepmann, N. A. Peppas. *Int. J. Pharm.***418**, 6 (2011).Search in Google Scholar

[44] J. H. Petropoulos, K. G. Papadokostaki, M. Sanopoulou. *Int. J. Pharm.***437**, 178 (2012).Search in Google Scholar

[45] P. Guru, A. K. Nayak, R. K. Sahu. *Colloids Surf., B***104**, 268 (2013).10.1016/j.colsurfb.2012.11.027Search in Google Scholar
PubMed

[46] S. Jana, A. Saha, A. K. Nayak, K. K. Sen, S. K. Basu. *Colloids Surf., B***105**, 303 (2013).10.1016/j.colsurfb.2013.01.013Search in Google Scholar
PubMed

[47] A. N. Shikov, O. N. Pozharitskaya, I. Miroshnyk, S. Mirza, I. N. Urakova, S. Hirsjärvi, V. G. Makarov, J. Heinämäki, J. Yliruusi, R. Hiltunen. *Int. J. Pharm.***377**, 148 (2009).Search in Google Scholar

[48] C. Li, C. Li, Y. Le, J.-F. Chen. *Int. J. Pharm.***404**, 257 (2011).Search in Google Scholar

[49] S. Setthacheewakul, S. Mahattanadul, N. Phadoongsombut, W. Pickayakorn, R. Wiwattanapatapee. *Eur. J. Pharm. Biopharm.***76**, 475 (2010).Search in Google Scholar

**Published Online:**2016-1-20

**Published in Print:**2016-2-1

©2016 IUPAC & De Gruyter. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. For more information, please visit: http://creativecommons.org/licenses/by-nc-nd/4.0/