- Email: [email protected]
Experimental Design for the Optimization of Lipid Nanoparticles
PHARMACEUTICAL NANOTECHNOLOGY Experimental Design for the Optimization of Lipid Nanoparticles J. ZHANG,1,2 Y. FAN,3 E. SMITH2 1
College of Pharmacy, University of Southern Nevada, 10920 S. River Front Parkway, South Jordan, Utah 84095
2
College of Pharmacy, University of South Carolina, 700 Sumter St., Columbia, South Carolina 29208
3
Department of Statistics, Case Western Reserve University, Cleveland, Ohio 44106
Received 20 March 2008; revised 3 June 2008; accepted 24 July 2008 Published online 9 September 2008 in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/jps.21549
ABSTRACT: Solid lipid nanoparticles (SLN) have been extensively investigated as a promising drug delivery system for controlling the release of therapeutic agents. Currently, there are many manufacturing methods available for SLN, including the high pressure homogenization method and the microemulsion technique. In addition, the solvent diffusion method has been discussed as an alternative technique in the literature, and has attracted great interest due to its simplicity and ease of handling. In order to gain a deeper understanding of this method, a statistical central composite design was applied in this study to examine how the physicochemical properties of the SLN were influenced by the variation of process parameters, including injected solvent, lipid concentration, surfactant concentration, temperature, and stirring speed. Our study showed that lipid concentration and temperature seemed to be the crucial parameters for the particle size of the monostearin SLN prepared by the solvent diffusion method. However, neither of these factors had a significant quadratic relationship with the zeta potential. ß 2008 Wiley-Liss, Inc. and the American Pharmacists Association J Pharm Sci 98:1813–1819, 2009
Keywords: experimental design; solid lipid nanoparticles; solvent diffusion; particle size; zeta potential
INTRODUCTION Solid lipid nanoparticles (SLN) have been used as a drug delivery system for controlling the release of various drugs.1 Several preparation techniques for SLN have been developed in the last decade, including high-pressure homogenization,2 microemulsion techniques,3 and solvent emulsification/ evaporation methods.4 The solvent diffusion Correspondence to: J. Zhang (Telephone: 801-8781077; Fax: 801-3020768; E-mail: [email protected]) Journal of Pharmaceutical Sciences, Vol. 98, 1813–1819 (2009) ß 2008 Wiley-Liss, Inc. and the American Pharmacists Association
method is another novel technique that has been reported in the literature recently for the preparation of SLN.5,6 This technique is commonly employed for the preparation of liposomes and polymer nanoparticles, and offers clear advantages over existing methods such as ease of handling and a fast production process, without the need for technically sophisticated equipment.7 To prepare SLN by the solvent diffusion method, the drug and lipid are dissolved in organic, water-miscible solvents at elevated temperatures and the resultant solutions are rapidly injected into an aqueous phase containing surfactants, under mechanical stirring.5,6 As the
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
1813
1814
ZHANG, FAN, AND SMITH
temperature cools, the lipid droplets solidify and thus SLN suspensions of the active drugs form. During the manufacturing process, many parameters appear to have a marked influence on the physicochemical properties of SLN. It is, therefore, essential to have a clear understanding of how preparation conditions determine particle characteristics and, in particular, how these characteristics are influenced by potential interactions between variables in the preparation process. Although the effects of preparation variables on the physicochemical properties have been reported,7 a systematic investigation of the simultaneous influence of multiple formulation variables on the SLN properties has not yet been undertaken. Evaluating the effect of a large number of formulation variables usually requires many experiments, which are often costly and time consuming. It is therefore prudent to minimize the total number of experiments performed in the optimization process, without sacrificing final product quality. Central composite design (CCD) has been used widely to find the operating parameters that optimize a specific manufacturing process.8 CCD is a collection of mathematical and statistical techniques that are useful for the modeling and analysis of problems composed of matrices of variables.9,10 Essentially, a relatively small number of empirical evaluations are used to determine mathematical trends that allow the prediction of final process parameters needed for a specific, optimized outcome. The CCD evaluation efficiently provides information on how the response of interest is influenced by several variables. This statistical tool significantly reduces the number of empirical experiments that are necessary to identify a mathematical trend in the experimental design, facilitating determination of the optimum level of variable factors required for a given response or result.11 The application of CCD is very flexible in design, allowing process evaluation under different experimental conditions of interest and operability. Therefore, the aims of the study were to evaluate the potential of a high throughput statistical screening method for theoretically optimizing the unit process parameters influential in the preparation of SLN, and to investigate the practical success of the predictive statistical process in the production of optimized SLN vehicles without the need for extensive empirical evaluation of the complex myriad of process parameters. JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
MATERIALS AND METHODS Materials Lecithin, poloxamer 188, tween 80, Sodium Dodecyl Sulfate, benzalkonium chloride, acetone, ethanol, and isopropanol were purchased from Fisher Scientific (Fairlawn, NJ). Monostearin (glyceryl monostearate) was purchased from Spectrum Chemical Mfg. Corp. (New Brunswick, NJ). Double-distilled, deionized water for HPLC was used throughout this study.
Methods Preparation of SLN The monostearin was completely dissolved in 5 mL of a water-miscible solvent in a water bath at 708C. Three solvents were used for screening purposes: Ethanol, isopropanol, and acetone. The resultant organic solution was injected into 50 mL of an aqueous phase containing surfactant at different temperatures, under mechanical agitation for 30 min. The nanosuspensions or nanoemulsions formed were then cooled to room temperature, resulting in SLN formations. The specific amount of ingredients added during the preparation strictly followed the experimental design conditions in Table 1. Characterization of Physicochemical Properties The particle size, polydispersity index, and electrophoretic mobility (zeta potential) of manufactured SLN dispersions were measured by Zetapals (Zetapals, Brookhaven Instruments, Holtsville, NY). Each sample was diluted with filtered, double-distilled water until an appropriate concentration of particles was achieved to avoid multiscattering events when measuring for 10 min in serial mode, with a sample time of 30 ms. The electrophoretic mobility was converted to a
Table 1. Factor Level Applied in the Optimization Factor Level in Design Factor X1 X2 X3 X4
2
1
0
þ1
þ2
lipid (mg/mL) 1 7 13 19 25 surfactant (%) 0 0.5 1 1.5 2.0 stirring speed (rpm) 100 300 500 700 900 temperature (8C) 20 35 50 65 80 DOI 10.1002/jps
EXPERIMENTAL DESIGN FOR THE OPTIMIZATION OF LIPID NANOPARTICLES
zeta potential using the Helmholtz–Smoluchowski equation by the software included within the Zetapals system. Experimental Design In this study, the preparation factors investigated were type of solvent, type of surfactant, concentration of surfactant, concentration of lipid, temperature, and stirring speed. Two qualitative factors (type of solvent and type of surfactant) were prescreened to simplify the experimental design and concentrate attention on optimization of the quantitative factors. Therefore only four quantitative factors were used for constructing the CCD, and the experimental conditions of the four factors at each level in our study are summarized in Table 1. A statistical software ‘‘MINITAB’’ was used to generate the representative combinations of these factors at different levels and the entire design consisted of 31 runs of experiments with three replicates of each set of conditions. Statistical Analysis Two optimal experimental responses were studied: Y1 ¼ Particle size; Y2 ¼ Zeta potential. They were the results of the individual influence and the interactions of the four independent variables. Hence the responses were modeled by the following polynomial model: Y ¼ b0 þ b1 X1 þ b2 X2 þ b3 X3 þ b4 X4 þ b11 X12 þ b22 X22 þ b33 X32 þ b44 X42 þ b12 X1 X2 þ b13 X1 X3 þ b14 X1 X4 þ b23 X2 X3 þ b24 X2 X4 þ b34 X3 X4 where the bi’s (for i ¼ 1, 2, 3, and 4) are the linear effects, the bii’s are the quadratic effects, the bij’s (for i, j ¼ 1, 2, 3, and 4, i < j) are the interaction between the ith and the jth variables; the b0 is the intercept. The results of these experiments were compared using analysis of variance (ANOVA), which was able to determine if the factors and the interactions between factors were significant. To test whether the terms were statistically significant in the regression model, t-tests were performed using a 95% (a ¼ 0.05) level of significance. An F-test was used to determine whether there was an overall regression relationship between the response variable Y and the entire set of X variables at a 95% (a ¼ 0.05) level of significance. DOI 10.1002/jps
1815
The coefficient of multiple determinations was denoted by R2, which measured the proportionate reduction of total variation in Y associated with the use of the set of X variables. In addition, the validity of the regression model was assessed according to statistical assumptions and lack of fit test. The statistical analysis was performed using MINITAB (Version 14). Model optimal operating conditions regarding the minimum or maximum response were solved and plotted by Maple (Version 9.5).
RESULTS AND DISCUSSION Preliminary Experiments Before the statistical experimental design was constructed, two qualitative factors (type of solvent and type of surfactant), were prescreened by varying only one factor at a time. Since the influence of each factor on the physicochemical properties of SLN was not known when the prescreening study was conducted, the experimental condition was set arbitrarily as follows: Surfactant concentration at 0.5%, monostearin concentration at 10 mg/mL, temperature at 208C, and stirring speed at 300 rpm. Several water-miscible solvents were screened for applicability in the preparation of SLN by the solvent diffusion method, including ethanol, isopropanol, and acetone. 0.5% Lecithin was used as the model surfactant. Methanol was excluded in this study as the solubility of monostearin in methanol was much lower than other solvents selected and also due to the toxicity of this solvent. The results are shown in Table 2. For the chosen conditions, the particle size of monostearin SLN obtained was the lowest using ethanol compared to acetone and isopropanol. The use of acetone as the organic solvent generated significantly larger nanoparticles with a particle size of more than 200 nm. Therefore, ethanol was selected as the model solvent for further investigation. However, different experimental conditions such as types of Table 2. Influence of Different Organic Solvents on the Monostearin SLN (n ¼ 3)
Solvent
Particle Polydispersity Zeta Size (nm) Index Potential (mV)
Ethanol 135.7 1.5 0.232 0.024 Isopropanol 177.5 1.5 0.221 0.027 Acetone 262.8 1.0 0.225 0.022
71.22 1.40 62.40 2.28 70.11 1.60
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
1816
ZHANG, FAN, AND SMITH
lipids and surfactants could lead to distinctively different results. In a similar study conducted by Schubert and Muller-Goymann,7 isopropanol seemed to be the best solvent in terms of particle size and size distribution, for the preparation of softisan1 100 SLN using 0.1% polysorbate 80 as the surfactant. The presence of surfactants reduces surface tension between the lipid and water and facilitates the solid particle formation during the cooling phase of SLN preparation. Several different types of surfactants were tested in terms of the particle size and zeta potential produced. The test list of surfactants included cationic, anionic, amphoteric, and nonionic surfactants. Ethanol was used as the model solvent in this screening. The results are shown in Table 3. In agreement with the literature,12 the particle size of the SLN dispersion produced with ionic and amphoteric surfactants was considerably smaller than that produced with nonionic surfactants. The polydispersity for all the SLN formulations with different surfactants were within the acceptable range (0–0.3), indicating homogeneous particle size distribution. The zeta potential of the SLN were all negative except the SLN produced by the cationic surfactant benzalkonium chloride, which was not stable and resulted in flocculation taking place within a few hours. The zeta potential of the SLN dispersion produced with nonionic surfactants was less negative than 30 mV, therefore the selected nonionic surfactants were not able to stabilize the SLN by simple electrostatic stabilization. The zeta potential of SLN using lecithin was the highest which indicated the best physical stability among these SLN produced by the different surfactants. In addition, lecithin is a relatively nonirritating surfactant compared with ionic surfactants, a distinct advantage for drug delivery systems. Therefore lecithin was selected as the model surfactant for further investigation. Strictly speaking, however, it is unwise to draw any conclusions from this preliminary screening study because there is certainly a risk of error in these interpolations and even more in extrapola-
tions. Nevertheless, this step is still necessary in order to simplify the problem and thus enable us to concentrate our attention and resources in a more detailed examination and optimization of several quantitative factors.11
Central Composite Design Effect of Formulation Variables on Particle Size The fitted least-squares, second-order equation for particle size Y1 is as follows: Y^1 ¼ 143 þ 33:1X1 þ 14:1X2 þ 1:70X3 27:2X4 þ 59:7X12 þ 3:90X22 þ 2:65X32 þ 19:1X42 6:28X1 X2 þ 15:2X1 X3 þ 3:64X1 X4 þ 11:2X2 X3 2:05X2 X4 þ 1:73X3 X4 The quadratic model was found to be significant with an F value of 14.65 ( p < 0.0001), which indicated that response variable Y1 and the set of X variables were significantly related. The high R2 value indicated that 93.2% of variation in particle size was explained by the regression on preparation factors. This model has a stationary point that is the minimum value, since all the quadratic regression coefficients are positive. The minimum particle size and its corresponding experimental setting were solved from the regression model. The minimum particle size of 142 nm could be achieved by operating the experiment under following experimental conditions: Monostearin concentration ¼ 13 mg/mL, surfactant concentration of 1.0%, stirring speed of 126 rpm, and temperature ¼ 61.58C. A verification test was conducted to prove the accuracy and usefulness of this statistical model under the optimized experimental conditions. The particle sizes of SLN formulations prepared under these conditions were analyzed (n ¼ 6) and compared with the predicted values of this model using a one sample t-test. The result showed that there was no statistically significant difference between the predicted and observed values ( p ¼ 0.2528).
Table 3. Influence of Different Surfactants on the Monostearin SLN (n ¼ 3) Surfactant Poloxamer 188 Tween 80 Lecithin SDS Benzalkonium chloride
Particle Size (nm)
Polydispersity Index
Zeta Potential (mV)
235.4 4.1 223.6 3.4 189.5 3.0 182.8 7.1 159.9 0.9
0.215 0.282 0.254 0.276 0.214
22.12 0.38 20.71 0.60 66.43 1.14 34.28 2.56 þ54.94 0.95
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
DOI 10.1002/jps
EXPERIMENTAL DESIGN FOR THE OPTIMIZATION OF LIPID NANOPARTICLES
The quadratic regression coefficients b11 and b44 were statistically significant while the other two coefficients were not. This may indicate that lipid concentration and temperature significantly influenced the particle size, whereas surfactant concentration and stirring speed had little impact on the particle size. The linear regression coefficients further confirmed the surfactant concentration and stirring speed did not change the particle size. Similarly, the cross term regression coefficients were not significant with the exception of b13, therefore almost no interactions were among experimental factors. The surface response plots for particle size as a function of preparation factors were constructed by fixing two of the variables at the optimized values. The most dramatic curvature was observed in Figure 1a in which particle size
1817
changes steeply by varying lipid concentration, but changes in a relatively gradual fashion viewed from the temperature axis. The lipid concentration seems to be the most important factor influencing the particle size. Schubert and Muller-Goymann7 reported that an increase of lipid concentration led to a concentration-dependent increase in particle size once the lipid concentration exceeded a critical concentration, which seems to be consistent with this study. The lipid particles produced with higher concentration than the experimental concentration may even lead to the formation of microparticles. The underlying mechanism can be explained by the change of diffusion rate of the organic solvent through the interface, which is a critical parameter for the formation of SLN prepared by the solvent diffusion method. When the concentration of
Figure 1. Estimated surface responses for particle size of SLN as a function of (a) lipid concentration and temperature, (b) surfactant concentration and temperature, (c) surfactant concentration and stirring speed. DOI 10.1002/jps
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
1818
ZHANG, FAN, AND SMITH
monostearin rises, the viscosity of the monostearin–ethanol diffusion phase also increases, which reduces the diffusion rates of the solute molecules. Surprisingly this study also showed that particle size did not necessarily always increase with lipid concentration. In fact, the particle size reaches a minimum point when the lipid concentration is 13 mg/mL in this model. The reason why the minimum particle size was observed during the study is still unclear. One possibility is that smaller particle size of the SLN may require higher surface energy thus higher thermodynamic activity and formulation instability.13 However, when the lipid concentration exceeds a limit, the major mechanism affecting the particle size may be the solvent diffusion rate and the collision and aggregation of the nanoparticles facilitated by high lipid concentration. The second most important factor is temperature as seen in Figure 1b, where particle size changes to a greater extent with temperature than with surfactant concentration. The minimum particle size could be achieved at 61.58C, which is close to the melting point (58–598C) of monostearin. When the temperature of the aqueous media is below the melting point, the lipid solidifies immediately when the solvent is injected to the aqueous media. Therefore, a higher temperature results in a greater molecular mobility rate and consequently a smaller particle size. However, when the temperature is above the melting point, the nanoemulsions may be initially formed before the cooling of the formulation. The high temperature may actually disturb the association of surfactant with the lipid and accelerate the collision of nanoemulsion droplets to form larger lipid particles during the cooling
process.14 Therefore the particle size seems to increase as the temperature increases. Figure 1c does not show a significant curvature when surfactant concentration and stirring speed are compared, indicating these factors are of minor influence on the resultant particle size. This suggests that in the range of lecithin concentration of 0.5–2%, there was no statistical difference in terms of stabilization of the newly generated surfaces and particle aggregation prevention. However, the existence of surfactant is necessary since no SLN could be formed in the absence of surfactant. On the other hand, increased stirring speed may not enhance the solvent diffusion rate as the particle size did not change significantly with stirring speed.
Effect of Formulation Variables on Zeta Potential The fitted least-squares second order equation for zeta potential Y2 is as follows: Y^2 ¼ 67:4 þ 4:49X1 4:19X2 þ 3:01X3 1:75X4 0:01X12 þ 1:06X22 1:12X32 þ 0:64X42 3:48X1 X2 þ 1:96X1 X3 þ 4:83X1 X4 0:69X2 X3 þ 0:36X2 X4 þ 0:26X3 X4 The quadratic model was found to be significant with an F value of 3.91 ( p ¼ 0.006), which indicated that response variable Y2 and the set of X variables were significantly related. However, the variation of zeta potential may not be well predicted by preparation factors because of the low R2 value of 78.5%. The quadratic regression coefficients were all statistically insignificant,
Figure 2. Estimated surface responses for zeta potential of SLN as a function of (a) lipid concentration and surfactant concentration, (b) stirring speed and temperature. JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
DOI 10.1002/jps
EXPERIMENTAL DESIGN FOR THE OPTIMIZATION OF LIPID NANOPARTICLES
which may indicate that zeta potential did not have quadratic relationships with the preparation factors. This suggests that the zeta potential, as an indication of the physical stability of the SLN, was not significantly affected by the experimental conditions employed in this study. Nevertheless, some of the linear and cross term coefficients are significant indicating that there might be some kind of relationship between zeta potential and experimental factors. No minimum zeta potential value can be achieved by manipulating the independent variables in this model. The surface response plots for zeta potential as a function of preparation factors (Fig. 2a and b) were constructed by fixing two variables at the optimized values obtained in the particle size model. No significant curvature could be observed by any two factors. The zeta potential was 78.3 mV by fitting the model with the optimized values obtained by the quadratic model for particle size, which is much higher than the mean value, indicating a formation of stable dispersion system (Mean ¼ 68.4303 mV, standard deviation ¼ 8.75576 mV). Therefore, the experimental conditions optimized by particle size models have been utilized empirically for preparation of SLN for other studies in our laboratory.
CONCLUSIONS This study demonstrates that the physicochemical properties of SLN prepared by the solvent diffusion method can be manipulated by variation of process parameters. The influence of the multiple preparation factors was studied and optimized with a surface response design using only a truncated set of empirical experiments and the statistical modeling of this empirical data was shown to be appropriate by checking the assumptions. The result of this mathematical analysis showed that lipid concentration and temperature seemed to be the crucial parameters for the particle size of the nanoparticles prepared by the solvent diffusion method; however none of the factors investigated in this study had a significant quadratic relationship with the zeta potential. Therefore, the statistical experimental design methodology has clearly shown its usefulness in this optimization process and this research serves as the groundwork for the understanding of SLN formation.
DOI 10.1002/jps
1819
REFERENCES 1. Muller RH, Mader K, Gohla S. 2000. Solid lipid nanoparticles (SLN) for controlled drug delivery—A review of the state of the art. Eur J Pharm Sci 50:161–177. 2. Almeida AJ, Runge S, Muller RH. 2000. Peptideloaded solid lipid nanoparticles (SLN): Influence of production parameters. Eur J Pharm Sci 50:161–177. 3. Gasco MR. 1993. Method for producing solid lipid microspheres having a narrow size distribution, US Patent No. 5250236. 4. Siekmann B, Westesen K. 1996. Investigation on solid lipid nanoparticles prepared by precipitation in o/w emulsion. Eur J Pharm Biopharm 43:104–109. 5. Hu FQ, Yuan H, Zhang HH, Fang M. 2002. Preparation of solid lipid nanoparticles with clobetasol propionate by a novel solvent diffusion method in aqueous system and physicochemical characterization. Int J Pharm 239:21–128. 6. Hu FQ, Jiang SP, Du YZ. 2005. Preparation and characterization of stearic acid nanostructured lipid carriers by solvent diffusion method in an aqueous system. Colloids Surf B Biointerfaces 45:167–173. 7. Schubert MA, Muller-Goymann CC. Solvent injection as a new approach for manufacturing lipid nanoparticles—Evaluation of the method and process parameters. Eur J Pharm Sci 55:125–131. 8. Box GEP, Hunter WG, Hunter JS, Hunter WG. 1978. Statistics for experimenters: An introduction to design, data analysis, and model building. New York: Wiley. 9. Julienne MC, Alonso MJ, Gomez Amoza JL, Benoit JP. 1992. Preparation of poly(D,L-lactide/glycolide) nanoparticles of controlled particle size distribution: Application of experimental designs. Drug Dev Ind Pharm 18:1063–1077. 10. Molpeceres J, Guzmam M, Aberturas MR, Chacon M, Berges L. 1996. Application of central composite designs to the preparation of polycaprolactone nanoparticles by solvent displacement. J Pharm Sci 85:206–213. 11. Lewis GA, Mathieu D, Phan-Tan-Luu R. 1999. Pharmaceutical experimental design. New York: Marcel Dekker, Inc. pp 186–191. 12. Mehnert W, Mader K. 2001. Solid lipid nanoparticles: Production, characterization and applications. Adv Drug Deliv Rev 47:165–196. 13. Byrappa K, Ohachi T. 2003. Crystal growth technology. Berlin, Germany: Springer. pp 325. 14. Freitas C, Muller RH. 1998. Effect of light and temperature on zeta potential and physical stability in solid lipid nanoparticle (SLN) dispersions. Int J Pharm 168:221–229.
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
College of Pharmacy, University of Southern Nevada, 10920 S. River Front Parkway, South Jordan, Utah 84095
2
College of Pharmacy, University of South Carolina, 700 Sumter St., Columbia, South Carolina 29208
3
Department of Statistics, Case Western Reserve University, Cleveland, Ohio 44106
Received 20 March 2008; revised 3 June 2008; accepted 24 July 2008 Published online 9 September 2008 in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/jps.21549
ABSTRACT: Solid lipid nanoparticles (SLN) have been extensively investigated as a promising drug delivery system for controlling the release of therapeutic agents. Currently, there are many manufacturing methods available for SLN, including the high pressure homogenization method and the microemulsion technique. In addition, the solvent diffusion method has been discussed as an alternative technique in the literature, and has attracted great interest due to its simplicity and ease of handling. In order to gain a deeper understanding of this method, a statistical central composite design was applied in this study to examine how the physicochemical properties of the SLN were influenced by the variation of process parameters, including injected solvent, lipid concentration, surfactant concentration, temperature, and stirring speed. Our study showed that lipid concentration and temperature seemed to be the crucial parameters for the particle size of the monostearin SLN prepared by the solvent diffusion method. However, neither of these factors had a significant quadratic relationship with the zeta potential. ß 2008 Wiley-Liss, Inc. and the American Pharmacists Association J Pharm Sci 98:1813–1819, 2009
Keywords: experimental design; solid lipid nanoparticles; solvent diffusion; particle size; zeta potential
INTRODUCTION Solid lipid nanoparticles (SLN) have been used as a drug delivery system for controlling the release of various drugs.1 Several preparation techniques for SLN have been developed in the last decade, including high-pressure homogenization,2 microemulsion techniques,3 and solvent emulsification/ evaporation methods.4 The solvent diffusion Correspondence to: J. Zhang (Telephone: 801-8781077; Fax: 801-3020768; E-mail: [email protected]) Journal of Pharmaceutical Sciences, Vol. 98, 1813–1819 (2009) ß 2008 Wiley-Liss, Inc. and the American Pharmacists Association
method is another novel technique that has been reported in the literature recently for the preparation of SLN.5,6 This technique is commonly employed for the preparation of liposomes and polymer nanoparticles, and offers clear advantages over existing methods such as ease of handling and a fast production process, without the need for technically sophisticated equipment.7 To prepare SLN by the solvent diffusion method, the drug and lipid are dissolved in organic, water-miscible solvents at elevated temperatures and the resultant solutions are rapidly injected into an aqueous phase containing surfactants, under mechanical stirring.5,6 As the
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
1813
1814
ZHANG, FAN, AND SMITH
temperature cools, the lipid droplets solidify and thus SLN suspensions of the active drugs form. During the manufacturing process, many parameters appear to have a marked influence on the physicochemical properties of SLN. It is, therefore, essential to have a clear understanding of how preparation conditions determine particle characteristics and, in particular, how these characteristics are influenced by potential interactions between variables in the preparation process. Although the effects of preparation variables on the physicochemical properties have been reported,7 a systematic investigation of the simultaneous influence of multiple formulation variables on the SLN properties has not yet been undertaken. Evaluating the effect of a large number of formulation variables usually requires many experiments, which are often costly and time consuming. It is therefore prudent to minimize the total number of experiments performed in the optimization process, without sacrificing final product quality. Central composite design (CCD) has been used widely to find the operating parameters that optimize a specific manufacturing process.8 CCD is a collection of mathematical and statistical techniques that are useful for the modeling and analysis of problems composed of matrices of variables.9,10 Essentially, a relatively small number of empirical evaluations are used to determine mathematical trends that allow the prediction of final process parameters needed for a specific, optimized outcome. The CCD evaluation efficiently provides information on how the response of interest is influenced by several variables. This statistical tool significantly reduces the number of empirical experiments that are necessary to identify a mathematical trend in the experimental design, facilitating determination of the optimum level of variable factors required for a given response or result.11 The application of CCD is very flexible in design, allowing process evaluation under different experimental conditions of interest and operability. Therefore, the aims of the study were to evaluate the potential of a high throughput statistical screening method for theoretically optimizing the unit process parameters influential in the preparation of SLN, and to investigate the practical success of the predictive statistical process in the production of optimized SLN vehicles without the need for extensive empirical evaluation of the complex myriad of process parameters. JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
MATERIALS AND METHODS Materials Lecithin, poloxamer 188, tween 80, Sodium Dodecyl Sulfate, benzalkonium chloride, acetone, ethanol, and isopropanol were purchased from Fisher Scientific (Fairlawn, NJ). Monostearin (glyceryl monostearate) was purchased from Spectrum Chemical Mfg. Corp. (New Brunswick, NJ). Double-distilled, deionized water for HPLC was used throughout this study.
Methods Preparation of SLN The monostearin was completely dissolved in 5 mL of a water-miscible solvent in a water bath at 708C. Three solvents were used for screening purposes: Ethanol, isopropanol, and acetone. The resultant organic solution was injected into 50 mL of an aqueous phase containing surfactant at different temperatures, under mechanical agitation for 30 min. The nanosuspensions or nanoemulsions formed were then cooled to room temperature, resulting in SLN formations. The specific amount of ingredients added during the preparation strictly followed the experimental design conditions in Table 1. Characterization of Physicochemical Properties The particle size, polydispersity index, and electrophoretic mobility (zeta potential) of manufactured SLN dispersions were measured by Zetapals (Zetapals, Brookhaven Instruments, Holtsville, NY). Each sample was diluted with filtered, double-distilled water until an appropriate concentration of particles was achieved to avoid multiscattering events when measuring for 10 min in serial mode, with a sample time of 30 ms. The electrophoretic mobility was converted to a
Table 1. Factor Level Applied in the Optimization Factor Level in Design Factor X1 X2 X3 X4
2
1
0
þ1
þ2
lipid (mg/mL) 1 7 13 19 25 surfactant (%) 0 0.5 1 1.5 2.0 stirring speed (rpm) 100 300 500 700 900 temperature (8C) 20 35 50 65 80 DOI 10.1002/jps
EXPERIMENTAL DESIGN FOR THE OPTIMIZATION OF LIPID NANOPARTICLES
zeta potential using the Helmholtz–Smoluchowski equation by the software included within the Zetapals system. Experimental Design In this study, the preparation factors investigated were type of solvent, type of surfactant, concentration of surfactant, concentration of lipid, temperature, and stirring speed. Two qualitative factors (type of solvent and type of surfactant) were prescreened to simplify the experimental design and concentrate attention on optimization of the quantitative factors. Therefore only four quantitative factors were used for constructing the CCD, and the experimental conditions of the four factors at each level in our study are summarized in Table 1. A statistical software ‘‘MINITAB’’ was used to generate the representative combinations of these factors at different levels and the entire design consisted of 31 runs of experiments with three replicates of each set of conditions. Statistical Analysis Two optimal experimental responses were studied: Y1 ¼ Particle size; Y2 ¼ Zeta potential. They were the results of the individual influence and the interactions of the four independent variables. Hence the responses were modeled by the following polynomial model: Y ¼ b0 þ b1 X1 þ b2 X2 þ b3 X3 þ b4 X4 þ b11 X12 þ b22 X22 þ b33 X32 þ b44 X42 þ b12 X1 X2 þ b13 X1 X3 þ b14 X1 X4 þ b23 X2 X3 þ b24 X2 X4 þ b34 X3 X4 where the bi’s (for i ¼ 1, 2, 3, and 4) are the linear effects, the bii’s are the quadratic effects, the bij’s (for i, j ¼ 1, 2, 3, and 4, i < j) are the interaction between the ith and the jth variables; the b0 is the intercept. The results of these experiments were compared using analysis of variance (ANOVA), which was able to determine if the factors and the interactions between factors were significant. To test whether the terms were statistically significant in the regression model, t-tests were performed using a 95% (a ¼ 0.05) level of significance. An F-test was used to determine whether there was an overall regression relationship between the response variable Y and the entire set of X variables at a 95% (a ¼ 0.05) level of significance. DOI 10.1002/jps
1815
The coefficient of multiple determinations was denoted by R2, which measured the proportionate reduction of total variation in Y associated with the use of the set of X variables. In addition, the validity of the regression model was assessed according to statistical assumptions and lack of fit test. The statistical analysis was performed using MINITAB (Version 14). Model optimal operating conditions regarding the minimum or maximum response were solved and plotted by Maple (Version 9.5).
RESULTS AND DISCUSSION Preliminary Experiments Before the statistical experimental design was constructed, two qualitative factors (type of solvent and type of surfactant), were prescreened by varying only one factor at a time. Since the influence of each factor on the physicochemical properties of SLN was not known when the prescreening study was conducted, the experimental condition was set arbitrarily as follows: Surfactant concentration at 0.5%, monostearin concentration at 10 mg/mL, temperature at 208C, and stirring speed at 300 rpm. Several water-miscible solvents were screened for applicability in the preparation of SLN by the solvent diffusion method, including ethanol, isopropanol, and acetone. 0.5% Lecithin was used as the model surfactant. Methanol was excluded in this study as the solubility of monostearin in methanol was much lower than other solvents selected and also due to the toxicity of this solvent. The results are shown in Table 2. For the chosen conditions, the particle size of monostearin SLN obtained was the lowest using ethanol compared to acetone and isopropanol. The use of acetone as the organic solvent generated significantly larger nanoparticles with a particle size of more than 200 nm. Therefore, ethanol was selected as the model solvent for further investigation. However, different experimental conditions such as types of Table 2. Influence of Different Organic Solvents on the Monostearin SLN (n ¼ 3)
Solvent
Particle Polydispersity Zeta Size (nm) Index Potential (mV)
Ethanol 135.7 1.5 0.232 0.024 Isopropanol 177.5 1.5 0.221 0.027 Acetone 262.8 1.0 0.225 0.022
71.22 1.40 62.40 2.28 70.11 1.60
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
1816
ZHANG, FAN, AND SMITH
lipids and surfactants could lead to distinctively different results. In a similar study conducted by Schubert and Muller-Goymann,7 isopropanol seemed to be the best solvent in terms of particle size and size distribution, for the preparation of softisan1 100 SLN using 0.1% polysorbate 80 as the surfactant. The presence of surfactants reduces surface tension between the lipid and water and facilitates the solid particle formation during the cooling phase of SLN preparation. Several different types of surfactants were tested in terms of the particle size and zeta potential produced. The test list of surfactants included cationic, anionic, amphoteric, and nonionic surfactants. Ethanol was used as the model solvent in this screening. The results are shown in Table 3. In agreement with the literature,12 the particle size of the SLN dispersion produced with ionic and amphoteric surfactants was considerably smaller than that produced with nonionic surfactants. The polydispersity for all the SLN formulations with different surfactants were within the acceptable range (0–0.3), indicating homogeneous particle size distribution. The zeta potential of the SLN were all negative except the SLN produced by the cationic surfactant benzalkonium chloride, which was not stable and resulted in flocculation taking place within a few hours. The zeta potential of the SLN dispersion produced with nonionic surfactants was less negative than 30 mV, therefore the selected nonionic surfactants were not able to stabilize the SLN by simple electrostatic stabilization. The zeta potential of SLN using lecithin was the highest which indicated the best physical stability among these SLN produced by the different surfactants. In addition, lecithin is a relatively nonirritating surfactant compared with ionic surfactants, a distinct advantage for drug delivery systems. Therefore lecithin was selected as the model surfactant for further investigation. Strictly speaking, however, it is unwise to draw any conclusions from this preliminary screening study because there is certainly a risk of error in these interpolations and even more in extrapola-
tions. Nevertheless, this step is still necessary in order to simplify the problem and thus enable us to concentrate our attention and resources in a more detailed examination and optimization of several quantitative factors.11
Central Composite Design Effect of Formulation Variables on Particle Size The fitted least-squares, second-order equation for particle size Y1 is as follows: Y^1 ¼ 143 þ 33:1X1 þ 14:1X2 þ 1:70X3 27:2X4 þ 59:7X12 þ 3:90X22 þ 2:65X32 þ 19:1X42 6:28X1 X2 þ 15:2X1 X3 þ 3:64X1 X4 þ 11:2X2 X3 2:05X2 X4 þ 1:73X3 X4 The quadratic model was found to be significant with an F value of 14.65 ( p < 0.0001), which indicated that response variable Y1 and the set of X variables were significantly related. The high R2 value indicated that 93.2% of variation in particle size was explained by the regression on preparation factors. This model has a stationary point that is the minimum value, since all the quadratic regression coefficients are positive. The minimum particle size and its corresponding experimental setting were solved from the regression model. The minimum particle size of 142 nm could be achieved by operating the experiment under following experimental conditions: Monostearin concentration ¼ 13 mg/mL, surfactant concentration of 1.0%, stirring speed of 126 rpm, and temperature ¼ 61.58C. A verification test was conducted to prove the accuracy and usefulness of this statistical model under the optimized experimental conditions. The particle sizes of SLN formulations prepared under these conditions were analyzed (n ¼ 6) and compared with the predicted values of this model using a one sample t-test. The result showed that there was no statistically significant difference between the predicted and observed values ( p ¼ 0.2528).
Table 3. Influence of Different Surfactants on the Monostearin SLN (n ¼ 3) Surfactant Poloxamer 188 Tween 80 Lecithin SDS Benzalkonium chloride
Particle Size (nm)
Polydispersity Index
Zeta Potential (mV)
235.4 4.1 223.6 3.4 189.5 3.0 182.8 7.1 159.9 0.9
0.215 0.282 0.254 0.276 0.214
22.12 0.38 20.71 0.60 66.43 1.14 34.28 2.56 þ54.94 0.95
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
DOI 10.1002/jps
EXPERIMENTAL DESIGN FOR THE OPTIMIZATION OF LIPID NANOPARTICLES
The quadratic regression coefficients b11 and b44 were statistically significant while the other two coefficients were not. This may indicate that lipid concentration and temperature significantly influenced the particle size, whereas surfactant concentration and stirring speed had little impact on the particle size. The linear regression coefficients further confirmed the surfactant concentration and stirring speed did not change the particle size. Similarly, the cross term regression coefficients were not significant with the exception of b13, therefore almost no interactions were among experimental factors. The surface response plots for particle size as a function of preparation factors were constructed by fixing two of the variables at the optimized values. The most dramatic curvature was observed in Figure 1a in which particle size
1817
changes steeply by varying lipid concentration, but changes in a relatively gradual fashion viewed from the temperature axis. The lipid concentration seems to be the most important factor influencing the particle size. Schubert and Muller-Goymann7 reported that an increase of lipid concentration led to a concentration-dependent increase in particle size once the lipid concentration exceeded a critical concentration, which seems to be consistent with this study. The lipid particles produced with higher concentration than the experimental concentration may even lead to the formation of microparticles. The underlying mechanism can be explained by the change of diffusion rate of the organic solvent through the interface, which is a critical parameter for the formation of SLN prepared by the solvent diffusion method. When the concentration of
Figure 1. Estimated surface responses for particle size of SLN as a function of (a) lipid concentration and temperature, (b) surfactant concentration and temperature, (c) surfactant concentration and stirring speed. DOI 10.1002/jps
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
1818
ZHANG, FAN, AND SMITH
monostearin rises, the viscosity of the monostearin–ethanol diffusion phase also increases, which reduces the diffusion rates of the solute molecules. Surprisingly this study also showed that particle size did not necessarily always increase with lipid concentration. In fact, the particle size reaches a minimum point when the lipid concentration is 13 mg/mL in this model. The reason why the minimum particle size was observed during the study is still unclear. One possibility is that smaller particle size of the SLN may require higher surface energy thus higher thermodynamic activity and formulation instability.13 However, when the lipid concentration exceeds a limit, the major mechanism affecting the particle size may be the solvent diffusion rate and the collision and aggregation of the nanoparticles facilitated by high lipid concentration. The second most important factor is temperature as seen in Figure 1b, where particle size changes to a greater extent with temperature than with surfactant concentration. The minimum particle size could be achieved at 61.58C, which is close to the melting point (58–598C) of monostearin. When the temperature of the aqueous media is below the melting point, the lipid solidifies immediately when the solvent is injected to the aqueous media. Therefore, a higher temperature results in a greater molecular mobility rate and consequently a smaller particle size. However, when the temperature is above the melting point, the nanoemulsions may be initially formed before the cooling of the formulation. The high temperature may actually disturb the association of surfactant with the lipid and accelerate the collision of nanoemulsion droplets to form larger lipid particles during the cooling
process.14 Therefore the particle size seems to increase as the temperature increases. Figure 1c does not show a significant curvature when surfactant concentration and stirring speed are compared, indicating these factors are of minor influence on the resultant particle size. This suggests that in the range of lecithin concentration of 0.5–2%, there was no statistical difference in terms of stabilization of the newly generated surfaces and particle aggregation prevention. However, the existence of surfactant is necessary since no SLN could be formed in the absence of surfactant. On the other hand, increased stirring speed may not enhance the solvent diffusion rate as the particle size did not change significantly with stirring speed.
Effect of Formulation Variables on Zeta Potential The fitted least-squares second order equation for zeta potential Y2 is as follows: Y^2 ¼ 67:4 þ 4:49X1 4:19X2 þ 3:01X3 1:75X4 0:01X12 þ 1:06X22 1:12X32 þ 0:64X42 3:48X1 X2 þ 1:96X1 X3 þ 4:83X1 X4 0:69X2 X3 þ 0:36X2 X4 þ 0:26X3 X4 The quadratic model was found to be significant with an F value of 3.91 ( p ¼ 0.006), which indicated that response variable Y2 and the set of X variables were significantly related. However, the variation of zeta potential may not be well predicted by preparation factors because of the low R2 value of 78.5%. The quadratic regression coefficients were all statistically insignificant,
Figure 2. Estimated surface responses for zeta potential of SLN as a function of (a) lipid concentration and surfactant concentration, (b) stirring speed and temperature. JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009
DOI 10.1002/jps
EXPERIMENTAL DESIGN FOR THE OPTIMIZATION OF LIPID NANOPARTICLES
which may indicate that zeta potential did not have quadratic relationships with the preparation factors. This suggests that the zeta potential, as an indication of the physical stability of the SLN, was not significantly affected by the experimental conditions employed in this study. Nevertheless, some of the linear and cross term coefficients are significant indicating that there might be some kind of relationship between zeta potential and experimental factors. No minimum zeta potential value can be achieved by manipulating the independent variables in this model. The surface response plots for zeta potential as a function of preparation factors (Fig. 2a and b) were constructed by fixing two variables at the optimized values obtained in the particle size model. No significant curvature could be observed by any two factors. The zeta potential was 78.3 mV by fitting the model with the optimized values obtained by the quadratic model for particle size, which is much higher than the mean value, indicating a formation of stable dispersion system (Mean ¼ 68.4303 mV, standard deviation ¼ 8.75576 mV). Therefore, the experimental conditions optimized by particle size models have been utilized empirically for preparation of SLN for other studies in our laboratory.
CONCLUSIONS This study demonstrates that the physicochemical properties of SLN prepared by the solvent diffusion method can be manipulated by variation of process parameters. The influence of the multiple preparation factors was studied and optimized with a surface response design using only a truncated set of empirical experiments and the statistical modeling of this empirical data was shown to be appropriate by checking the assumptions. The result of this mathematical analysis showed that lipid concentration and temperature seemed to be the crucial parameters for the particle size of the nanoparticles prepared by the solvent diffusion method; however none of the factors investigated in this study had a significant quadratic relationship with the zeta potential. Therefore, the statistical experimental design methodology has clearly shown its usefulness in this optimization process and this research serves as the groundwork for the understanding of SLN formation.
DOI 10.1002/jps
1819
REFERENCES 1. Muller RH, Mader K, Gohla S. 2000. Solid lipid nanoparticles (SLN) for controlled drug delivery—A review of the state of the art. Eur J Pharm Sci 50:161–177. 2. Almeida AJ, Runge S, Muller RH. 2000. Peptideloaded solid lipid nanoparticles (SLN): Influence of production parameters. Eur J Pharm Sci 50:161–177. 3. Gasco MR. 1993. Method for producing solid lipid microspheres having a narrow size distribution, US Patent No. 5250236. 4. Siekmann B, Westesen K. 1996. Investigation on solid lipid nanoparticles prepared by precipitation in o/w emulsion. Eur J Pharm Biopharm 43:104–109. 5. Hu FQ, Yuan H, Zhang HH, Fang M. 2002. Preparation of solid lipid nanoparticles with clobetasol propionate by a novel solvent diffusion method in aqueous system and physicochemical characterization. Int J Pharm 239:21–128. 6. Hu FQ, Jiang SP, Du YZ. 2005. Preparation and characterization of stearic acid nanostructured lipid carriers by solvent diffusion method in an aqueous system. Colloids Surf B Biointerfaces 45:167–173. 7. Schubert MA, Muller-Goymann CC. Solvent injection as a new approach for manufacturing lipid nanoparticles—Evaluation of the method and process parameters. Eur J Pharm Sci 55:125–131. 8. Box GEP, Hunter WG, Hunter JS, Hunter WG. 1978. Statistics for experimenters: An introduction to design, data analysis, and model building. New York: Wiley. 9. Julienne MC, Alonso MJ, Gomez Amoza JL, Benoit JP. 1992. Preparation of poly(D,L-lactide/glycolide) nanoparticles of controlled particle size distribution: Application of experimental designs. Drug Dev Ind Pharm 18:1063–1077. 10. Molpeceres J, Guzmam M, Aberturas MR, Chacon M, Berges L. 1996. Application of central composite designs to the preparation of polycaprolactone nanoparticles by solvent displacement. J Pharm Sci 85:206–213. 11. Lewis GA, Mathieu D, Phan-Tan-Luu R. 1999. Pharmaceutical experimental design. New York: Marcel Dekker, Inc. pp 186–191. 12. Mehnert W, Mader K. 2001. Solid lipid nanoparticles: Production, characterization and applications. Adv Drug Deliv Rev 47:165–196. 13. Byrappa K, Ohachi T. 2003. Crystal growth technology. Berlin, Germany: Springer. pp 325. 14. Freitas C, Muller RH. 1998. Effect of light and temperature on zeta potential and physical stability in solid lipid nanoparticle (SLN) dispersions. Int J Pharm 168:221–229.
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 5, MAY 2009