Application of the Box–Behnken design for the production of soluble curcumin: Skimmed milk powder inclusion complex for improving the treatment of colorectal cancer

Abstract The main objective of this study was to develop a soluble product of the practically insoluble curcumin (CMN) to treat colorectal cancer more effectively than with pure CMN. To improve the solubility of CMN, various hydrophilic carriers of skimmed milk powder (SMP), polyvinylpyrrolidone (PVP), and mannitol (MNT) were utilized to prepare solid dispersion (SD) binary complexes. The prepared complexes were characterized in terms of their aqueous solubility and in vitro drug release and analyzed by Fourier transform infrared spectrophotometry, powder X‐ray diffractometry, scanning electron microscopy, dynamic light scattering, and the novel dyeing test. Based on this characterization, the best SD complex was optimized using the Box–Behnken design (RSM‐BBD). These results showed that the solubility of CMN was greatly improved in combination with SMP. The SD of CMN with SMP produced significantly improved solubility (0.646 ± 0.024 mg/ml) and dissolution (54.94 ± 3.21% at 5 min). Further, solid‐state characterization revealed that the complex exhibited intermolecular inclusion of the drug and carrier. Also, the complex did not undergo any chemical modification owing to its amorphous form, and the novel dye test showed better coloring impact, indicating the solubility of CMN. The in vitro cytotoxicity of the complex showed that 50% inhibition (IC50) of SW480 and Caco‐2 cells was achieved at a considerably lower concentration than that of pure CMN. Flow cytometry analysis confirmed that the cell cycle arrest was at G2/M phase (43.26% and 65.14%), and DNA fragmentation analysis investigation confirmed that the complex induced more DNA damage during apoptosis.


| INTRODUC TI ON
Skimmed milk powder (SMP) is prepared from pure milk by pasteurization followed by evaporation, and then, the powdered soluble SMP can be obtained using spray dryer. The composition is fat (1%), protein (35%), carbohydrates (51%), minerals (7%), and moisture (3.5%) ( Table 1). SMP is composed of alpha-, beta-, and kappa-casein and acts as an emulsifying agent for oral pharmaceutical preparations (Shah et al., 2009). SMP can enhance drug solubility due to the hydrophobic protein component, that is, casein being entrapped in to a lipophilic/hydrophobic drug, and the reaction (inclusion complex) is caused by β-casein acting as a surfactant. The drug is subsequently released due to the presence of kappa-casein on the surface of the hydrophilic inclusion complex; this improves the wettability and surface tension of the complex, leading to enhancement of solubility of the drug (Palanisamy & Khanam, 2010).
Most anticancer drugs can develop toxicity in the body.
Synthetic or semisynthetic drugs may have various side effects such as bone marrow depression, alopecia, extreme fatigue, loss of self-confidence, loss of immunity, and a remarkable decline in WBC count. The insolubility of anticancer drugs leads to their accumulation at the absorption site, and this may develop toxicity (Nurgali et al., 2018). The dosing frequency can be minimized by establishing the proper dosage form to deliver the drug over an extended period.
Numerous studies on curcumin (CMN) have proven that its principal active component-lipophilic polyphenol-produces various anticancer activity; however, its weak pharmacokinetics, including low solubility and stability as well as rapid metabolism and elimination, have limited its therapeutic application (Sachin et al., 2016).
The enhancement of the solubility of insoluble drugs is an important and one of the most challenging tasks in present-day research. The diverse physicochemical methodologies that have been explored to enhance drug solubility include the addition of surfactants, nanosizing and micronization, conversion to the amorphous state, increasing wettability, developing prodrugs and salts, the liposome approach (Bansal et al., 2011), and SD. The basic mechanism by which SD improves the solubility of insoluble powder is by blending a carrier and drug, thus completely altering the state from semisolid/liquid to solid. Finally, the resultant product is milled to the greatest extent possible and then suitably sieved (Seo et al., 2012).
The Box-Behnken design (BBD) is an efficient choice of response surface methodology (RSM; Otto, 1999). Each factor has three levels, that is, low, center, and high. It is rotatable per factor with these three levels and joins a fractional factorial with inadequate block plans to avoid high peaks. This design is strictly limited to conditions where the researcher is not concerned with finding the response at the angles of the cube (boundaries). BBD is commonly used than the Doehlert design because it requires fewer number of experiments (Hanrahan et al., 2005). It requires greater skill and is ultimately analyzed by separating the number of coefficients to the quadratic condition by the number of analyses necessary for the design.
In this study, increasing the solubility of CMN was explored by preparing complexes with solid dispersion (SD) carriers of SMP, polyvinylpyrrolidone (PVP), and mannitol (MNT).

| Phase solubility study
The phase solubility (PS) study was according to the methodology of Higuchi and Connors (1965 Here, symbols S and I denote the slope and intercept, respectively. Additionally, the enthalpy change (ΔH) on complexation was obtained from the Van't Hoff equation (Equation 2).
Here, K 2 , K 1 , and t 2 and t 1 refer to the stability constants and corresponding temperatures in Kelvin at 37 and 25°C, respectively. The Gibbs free energy change (ΔG) and change in entropy (ΔS) upon ideal complexation were computed from Equations 3 and 4, respectively.

| Preparation of PM and SD
Physical mixtures (PM) of varying compositions (1:3 − 1:7 of CMN to carrier, respectively) were prepared with a mortar and pestle as specified by the guidelines of geometrical mixing and by sieving (# 120; 150 − 125 µm). SD complexes of same compositions were prepared by the following methods. 100 mg of CMN was used in each preparation.

| Solvent evaporation method
CMN was dissolved in ethanol, followed by addition of carrier (SMP/ PVP). The resulting dispersion was stirred at 40 ± 0.5°C by magnetic stirrer (REMI-2MLH) and then transferred to a petri dish heated to 50 ± 0.5°C on a hot plate to effect evaporation of the solvent.
The obtained product was desiccated at 40°C for 48 hr (Verma et al., 2017).

| Melting method
The SD was prepared by adding CMN (100 mg) to the liquefied carrier (MNT) at 70°C with constant stirring at 700 rpm for 15 min until the product obtained as homogenous. The resulting product was allowed to solidify by cooling at 28°C, after which it was powdered, sieved, and stored in a desiccator (Vasoya et al., 2019).

| Aqueous solubility study
An excess quantity of sample (pure CMN, SD, or PM) was added to Milli Q water in a 100-ml volumetric flask and heated in a water bath at a steady temperature of 37 ± 1°C for 24 hr, with shaking at 30min intervals (Mohamed et al., 2019). Subsequently, the mixture was filtered through a 0.45-µm Millipore membrane filter and diluted to a suitable concentration and the UV absorbance was measured.

| Dyeing experiment
A simple novel dye test can be applied to determine the effective solubility of a sample (usually colored) in aqueous media (Lanxiang et al., 2017). Briefly, CMN (10 mg) and its SDs (equiv. wt.) were added to double distilled water (15 ml). The resulting mixture was sonicated for 5 min followed by filtration, and then, images of the solutions were taken. The above solution was then diluted to 50 ml, and six white linen cloths (8 × 4.5 cm 2 ) were soaked in it for 1.5 hr. The cloths were then dried at room temperature and observed for color intensity.

| Dissolution study
The dissolution study was performed in 900 ml of double distilled water at 37 ± 1°C at 50 rpm using a USP dissolution apparatus II (DS 8000, Lab India). The sample was placed into the jar and the time set as zero. At each 5-min interval until 30 min, 5 ml sample aliquots were syringed out and filtered through using 11-µm Whatman filter paper. The percentage of dissolved CMN was estimated by UV spectrometer. A correction for the collective dilution initiated by addition of the sample to the double distilled water was applied to maintain the sink condition.

| Optimization by BBD
The BBD is a dynamic second-order trial configuration related to the minimum trial experiments. BBD was employed for optimization of the SD complex preparation variables. The Design-Expert® version 12 software trial (Stat-Ease Inc.) was utilized for arithmetical exploration by ANOVA, enabling construction of model equations and 3D response plots for each outcome. Carrier concentration (X 1 ), stirring rate (X 2 ), and temperature (X 3 ), and the corresponding dependent variables (Y 1 and Y 2 ), aqueous solubility (Sol aq , mg/mL), and release in 5 min (Rel 5 min , %) are the factors that facilitate the preparation of powder complexes (Chaudhary et al., 2011). The quadratic RSM was fitted to the following Equation 5. In each preparation, 100 mg of CMN was used.

| Dynamic light scattering analysis (DLS)
The hydrodynamic particle size (PS), polydispersible index (PDI), and zeta potential (ZP) of the PM and SD complexes were determined by DLS (Mohamed et al., 2019) using a Zetasizer Nano ZS90 (Malvern Instruments Ltd.).

| Scanning electron microscopy (SEM)
The surface characteristics of samples were examined using SEM to observe electrons moving in zig-zag patterns (Carl Zeiss Microscopy Ltd, EVO 18). The test samples were diluted with Milli Q water (1:100), dropped onto two-sided carbon adhesive tape pre-affixed on a sample stub, and then allowed to dry at 28°C (Kumar et al., 2017).
A thin layer of sample was gold coated (100 Å) by a sputter coater, following which the samples were observed at 5.0 kV.

| MTT assay
The cytotoxicity of complexes against Caco-2 and SW480 cell lines was estimated by MTT assay (Dhananjay et al., 2014). Briefly, cells were seeded at a density of 5 × 10 3 cells·mL −1 (200 µl/well) in 96-well culture plates with DMSO as the solvent. After 24 hr of incubation with different concentrations of complex, MTT reagent (5 mg/ml, 20 µl/well) was added to the medium and incubated at 37°C for 4 hr.
The obtained purple formazan precipitate was solubilized by addition of DMSO (100 μl) to each well. The cell passage number was 3. The absorbance of each well was estimated at 570 nm by a plate reader (Bio-Rad, iMark). Experiments were performed in triplicate, and the mean calculated. The IC 50 value was determined as the concentration of complex required to decrease the absorbance to half the percentage of the control (Lakshmipraba et al., 2013).

| Acridine orange (AO) and ethidium bromide (EB) staining
The AO/EB dual stain method was used to investigate the apoptotic morphology, with some modifications. Briefly, the IC 50 concentra-

| Hoechst staining
The Caco-2 and SW480 cells were seeded in separate 6-well plates and treated using the IC 50 concentrations. After incubation for 24 hr, cells were collected and stained by aqueous Hoechst 33258 stain at room temperature for 5 min (Nagaraj et al., 2014). Morphological variations were also observed and micrographed by fluorescence microscopy with a 377−355 nm filter (400× magnification).

| DNA fragmentation assay
DNA fragmentation was performed using the agarose gel electrophoresis method. The SW480 and Caco-2 cells were cultured at 1 × 10 6 cells/ml in 6-well culture plates. Nearly 70% of the cells joined together, and the inclusion complex was added and incubated for 48 hr. After incubation, cells were collected and the pellets centrifuged (Eppendorf centrifuge 5430R) and washed twice with icecold PBS. The pellets were lysed by treating with EDTA (10 mM), Tris-HCl (10 mM), and Triton X-100 (0.2%; pH 7.5) in cold buffer for 10 min. The mixtures were then centrifuged at 4°C and 13,000 g for 10 min, and the supernatant containing fragmented DNA, RNA, and undamaged chromatin was treated with IA-P-C (1:24:25) (Dasiram et al., 2017). The pellet was washed with ethanol (70%), dried at room temperature, and dissolved in T10E1 buffer (adjusted to pH 7.4, 20 μL). Agarose (2%) gel electrophoresis was utilized to assay 0.6% mg/mL of RNase A at 37°C for 30 min. At the end point, ethidium bromide was used to stain the gels and the DNA fragment observed by UV transillumination.

| Flow cytometry
Caco-2 and SW480 cells at 1 × 10 5 cells/mL were incubated with the IC 50 concentration as determined above for 24 hr to analyze the ef-

| Outcomes of PS study
The PS study confirmed the proposed effect of the various carriers on improving the solubility of CMN. A standard linear curve was obtained in the concentration ranges from 5.97 × 10 −5 to 2.04 × 10 −4 mM and 8.14 × 10 −5 to 2.12 × 10 −4 mM for CMN at 25 and 37°C, respectively ( Table 2).
The casein)-which acts as an emulsifier and surface active agent-may lead to strong interaction with CMN as the hydrophobic hydrogen groups (Hartel & Hasenhuettl, 2008). Therefore, this result was obtained due to the establishment of a soluble complex between the polymeric carrier and lipophilic drug.
The CMN-PVP complex was practically unsuitable because of its narrow solubility characteristics. Hydrophobic interactions between the hydrogen bond of CMN with the carbonyl group of PVP were likely involved between CMN and PVP (Chatjigakis et al., 1992). This interaction may have impacted the entropy, with water molecules accumulating around exposed nonpolar solute particles; however, when two nonpolar molecules approached, the disorder of water molecules increased, causing a positive change to the entropy and free energy of association.
An exception was noticed for MNT where a high + ΔS value (4.4 J mol −1 K −1 ) was obtained due to ionization of ligand molecules, resulting in water molecules being less ordered. The binding process was endothermic (+ΔH); ΔH favors ΔG and ΔS, and the spontaneity ensured by negative ΔG meant that at least complex formation occurred between MNT and CMN.

| CMN solubility
The solubility of pure CMN in distilled water was found to be 0.004 mg/ml at 37°C after 24 hr. The solubility of CMN-SMP (1:3 to 1:7) SD was enhanced ~120-to 150-fold compared to that of pure CMN (Table 3). In contrast, the PVP and MNT binary systems showed less enhancement of solubility-25-to 30-fold and 16-to 23-fold, respectively-due to less interaction of these carriers with CMN (El-Badry, 2011). The amino acid content of SMP was primarily responsible for the greater solubility enhancement of the CMN-SMP complex.
Overall, the aqueous solubility of CMN in the SDs was enhanced to a greater degree-depending on the carrier concentration-than in the PMs. The aqueous solubility of the PMs was of less importance than the solubility of the SDs due to the partial or improper complexation of the carriers with CMN.  in water increased the most after incorporation in this complex (Xu et al., 2015).

| FT-IR
FT-IR spectra of pure CMN, the carriers, and the PM and SD com-

| PXRD
The PXRD patterns of CMN, the carriers, PMs, and SDs are shown in Figure 3b. CMN displayed sharp peaks (indicating a crystalline arrangement) at 2θ = 8.98º and 17.38º and many tiny peaks at 23.48º, 24.72º, 25.68º, 26.22º, and 27.5º (Jang et al., 2014). Many diffraction points with high intensity were observed for MNT, with strong and intense peaks at 2θ = 18.96, 21, and 23.54º. The same MNT X-ray diffraction pattern (Figure 3b) was observed in the PM. In contrast, the CMN peak at 2θ = 9.28° was decreased in intensity in the SD; this was because there were no changes in crystallinity, and the peaks of MNT remained unchanged, indicating that there was little interaction occured between the drug and this carrier.
In contrast, PVP and SMP showed amorphous characteristics because of their lack of comprehensive stereo uniformity and exist the large lateral groups in these carrier (Figure 3b). The obtained X-ray diffraction patterns revealed that this amorphous nature was fundamental in SMP-PM because its low intensity was due to the

| CMN release state
The average dissolution curves of CMN and its PM and SD are shown in Figure 4. It is notable that the dissolution rate of pure CMN was lowest (1.62%) at around 30 min due to its high hydrophobicity. The dissolution rate of the complexes of CMN and the carriers exhibited great burst release (12%-40%) in the initial 5−6 min (Table 4); this showed that the entire complex established with the carrier or was converted to an amorphous mass (Sadeghi et al., 2016).  (Li et al., 2015). Thus, SMP gave the best solubility up to a 1:7 ratio, at higher ratios of carrier the solubility declined, indicating that dissolution is dependent on carrier concentration. The in vitro release demonstrated that, after the burst release, a steady rate release profile was seen with all SDs (Table 4).

F I G U R E 4
Cumulative drug release of CMN from (a) PM and (b) SD (mean ± SD, n = 3) Rapid dissolution of CMN, slightly less than SD, may be clarified as an in situ complex improvement arising from decreased interfacial tension (Figure 4a). Remarkably, the hydrophilic carrier PVP increased the wettability and spreadability of CMN by lessening the amount of CMN that was immediately soluble. PVP is a hydrophilic carrier, and, the viscosity of the binary system increased, which in turn decreased the dispersion of the drug molecule and helped to form a crystal lattice. The CMN-SD obtained with MNT was not satisfactory in comparison to the other carriers; this was likely due to the polar effect of the carbohydrates and the formation of hydrogen bonds improving solubilization (Saharan et al., 2009). This result showed that the limited dissolution rate of MNT could due to its partial conversion from the crystalline to the amorphous form in the presence of pure CMN powder. As an outcome of these findings, the low dissolution rate of MNT may be a result of inadequate complexation or amorphization of the CMN particles, compared with those of pure CMN.

| Optimization of Variables
The experimental results showed that the CMN-SD SMP binary system prepared by the solvent evaporation method had enhanced solubility and dissolution of CMN, with better physicochemical

TA B L E 5 Effect of independent process variables on dependent variable by Box-Behnken design
Preparation code

Actual Predicted Actual Predicted
properties. Subsequently, BBD was employed for the optimization of the SMP SD complex. The carrier (SMP) concentration (X 1 ), stirring rate (X 2 ), and temperature (X 3 ) were the independent variables for the three-factor, two-level design, and the dependent variables selected were aqueous solubility (Y 1 ; Sol aq , mg/ml) and release in 5 min (Y 2 ; Rel 5 min ; Table 5).
The independent coded values of variables were as follows.

| Sol aq
The quadratic models were fitted for the dependent variables by the software. The polynomial model for a dependent variable Y 1 is denoted by the following equation (Equation 6): (6) Y 1 = +0.5460 + 0.13 * X 1 + 0.014 * X 2 − 0.020 * X 3 − 0.010 * X 1 * X 2 − 0.005 * X 1 * X 3 − 0.0010 * X 2 * X 3 − 0.0010 * X 2 * X 3 − 0.050 * X 2 1 − 0.041 * X 2 2 + 0.025 * X 2 3 . The interaction properties of the dynamic response were validated by graphical representation, which were dependent on the CC (X 1 ), SR (X 2 ), and Temp (X 3 ), and the resulting aqueous solubility of CMN (Y 1 ), as shown in Figure 5a. An increase in the concentration of SMP from 500 to 1,000 mg significantly enhanced the solubility of CMN from 0.34 to 0.76 mg/ml. This revealed that the carrier was effectively responsible for the aqueous solubility of the drug. On the other hand, the aqueous solubility of CMN slightly decreased from 70 to 40°C. The SR showed that a decrease in rpm (1,000 to 500) enhanced the aqueous solubility of CMN, although this effect was not prominent (Baig et al., 2016).

| Rel 5min
The following equation (Equation 7) represents the polynomial model for the response (release at 5 min; Y 2 ): Of all of the independent variables, CC was observed to have a significantly lower positive result on the release at 5 min as was manifest in the narrow positive value for its coefficient. The negative values for the SR and Temp showed that the release decreased at a higher rate of stirring and temperature.
The negative effect of increasing the stirring rate was found to be greater than the effect of increasing the Temp. The interaction between the independent variables was also found to be significant. Overall, the model was typical (p < .0001; F-value = 102.46), while the lack of fit was not significant (p = .67; F-value = 0.6115).
The predicted value (.9517) and adj.R 2 (.9828) values are important parameters (Chaudhary et al., 2011). The CC significantly affected the release of CMN from SD at the end of 5 min: If the CC was too small, release from SD could not be achieved at the desired level; when CC was 1,000 mg, the release was up to 50%−55% (Figure 5b).
Therefore, a suitable CC was selected for the release of CMN from SD. These results indicated that the release rate of CMN increased rapidly by increasing the SR and decreasing the temperature; these effects were significant for CC of 500 and 1,000 mg, and the center point of SR and Temp revealed that the release of CMN significantly increased as the temperature decreased, which could be due to complete amorphization of CMN.
The optimized formulation (P18) validated for the response variables (Sol aq and Rel 5min ) was set at a maximum range, and the optimized preparation produced by the design noted as 850 mg of SMP, 750 rpm SR, and 40°C Temp. Accordingly, CMN-SMP SD was prepared by the solvent evaporation method with these specifications. The experimental values of Sol aq and Rel 5min (0.646 mg/ml and 54.94%, respectively) were compared with the predicted values (0.632 mg/ml and 55.71%), and the prediction error was found to be 1.02 and 0.98%, respectively. The negligible errors confirmed that the established model was sound, and the results of the predicted responses were in accordance with the investigational data.

| DLS
The hydrodynamic particle size of the PM was 383.4 nm and the PDI was 0.497 ( Figure 6). This confirmed the bimodal size distribution, from which it was evident that the PM solution consisted of at least two constituents (Kumar et al., 2016). The hydrodynamic diameter of SD was 275.3 nm with a PDI of 0.239 (Figure 6b). The low PDI indicated the homogeneity of particle size distribution for the SD complex with CMN.
ZP is a significant parameter for understanding the state of the particle surface charge and stability (long-term storage) of the products (Pooja et al., 2015). The ZP of the CMN-PM and CMN-SD complexes was found to be −23.2 and −33.9 mV, respectively ( Figure 6c and d), which indicated that the powdered SD was molecularly dispersed with CMN to give a superior inclusion complex than the PM. In contrast, in the SD image ( Figure 7d) the drug surface appeared to be more porous, uniform and homogeneously dispersed.

| SEM
The morphology of both the CMN and SMP had disappeared at the molecular level, revealing their complete conversion into the molecular level inclusion form.

| In vitro cytotoxicity of SMP inclusion complex
The cytotoxic outcomes of the MTT reduction assay of pure CMN and optimized CMN-SMP inclusion complex on Caco-2 ( Figure 8a) and SW480 and cells lines are shown in Figure 8b. The IC 50 values for the complexes were calculated as being around 73 and 89 µM/ ml for SW480 and Caco-2, respectively, whereas pure CMN was 116 and 132 µM/ml (Sun et al., 2013).
The outcomes of the MTT assay showed that the inclusion complex exhibited better cytotoxicity than pure CMN. This could be due to variations in cellular uptake leading to the improved effect of the complex (Manju & Sreenivasan, 2012). The effects of this assay indicated that the complex could transport the drug to Caco-2 and SW480 cells using endocytic process by active targeting combined with better cytotoxicity and aqueous solubility of CMN. The cell viability gradually declined with increasing concentration of C-SD and pure CMN, while SD (soluble CMN) showed a significantly better cytotoxic effect than pure CMN.

| Hoechst staining
Cytological changes of the SW480 and Caco-2 cells were detected after treatment with the IC 50 concentrations of the complexes for 24 hr (Vignesh et al., 2014). A manual count of normal and abnormal cells was performed using Hoechst 33258 staining of the Sw480 and Caco-2 cells, shown in Figure 9c,d (Ramakrishnan et al., 2011).
This method showed abnormalities in the cell cytology, with specific reference to the nucleus core and cytoplasm at the primary

| DNA fragmentation
To trigger apoptosis in CMN-treated Caco-2 and SW480 cells, a DNA fragmentation assay was performed using the agarose gel electrophoresis method. This reveals "fragment or ladder" pattern of DNA and  (Vignesh et al., 2014).
An in-depth visualization displayed the outstanding outcome on the initiation of apoptosis as evidenced by the presence of DNA ladders in the samples treated with inclusion complex, exhibiting more active cell death compared with pure CMN (more cleavage of the DNA "ladder").

| Cell cycle analysis
Propidium iodide staining was used to enable morphological inves-  paths (Oberhammer et al., 1993), which activated the apoptotic pathway subsequently treated with SMP inclusion complex.

| CON CLUS ION
In conclusion, this study determined that the ideal stoichiometric ratio of CMN with carrier to enhance its solubility was 1:1. This complex was found to be more effective against colorectal cancer (CRC) than pure CMN. The solubility and dissolution proper-